Showing 1–50 of 145 results for author: Mitchell, M

High-Performance Reinforcement-Learned BP Decoding of Quantum LDPC Codes

Authors: Mohsen Moradi, Vahid Nourozi, Taejoon Kim, Remi A. Chou, David G. M. Mitchell

Abstract: Belief-propagation (BP) decoding is attractive for quantum low-density parity-check (QLDPC) codes because it uses local message passing on sparse Tanner graphs. However, conventional flooding BP often stalls due to stabilizer degeneracy and short cycles. Reinforcement-learning-based sequential variable-node scheduling (RL-S), which learns the update order offline, has shown that adaptive schedulin… ▽ More Belief-propagation (BP) decoding is attractive for quantum low-density parity-check (QLDPC) codes because it uses local message passing on sparse Tanner graphs. However, conventional flooding BP often stalls due to stabilizer degeneracy and short cycles. Reinforcement-learning-based sequential variable-node scheduling (RL-S), which learns the update order offline, has shown that adaptive scheduling can improve BP convergence. In this paper, we extend this idea with a second-order local update decoder, RL-S2LU. The proposed decoder preserves BP locality and low complexity, while numerical results show significant error-correction gains over conventional BP and the considered BP-OSD-10 baseline. △ Less

Submitted 27 July, 2026; originally announced July 2026.

Learning to Decode Quantum LDPC Codes via Cluster-Based Sequential Belief Propagation

Authors: Mohsen Moradi, Taejoon Kim, Rémi A. Chou, David G. M. Mitchell

Abstract: Belief-propagation (BP) decoding for quantum low-density parity-check (QLDPC) codes is attractive due to its low complexity, but its performance is often limited by short cycles, degeneracy, and convergence failures. Recently, reinforcement-learning-based sequential variable-node (VN) scheduling (RL-S) was shown to improve BP decoding by learning state-dependent update orders. However, the VN-by-V… ▽ More Belief-propagation (BP) decoding for quantum low-density parity-check (QLDPC) codes is attractive due to its low complexity, but its performance is often limited by short cycles, degeneracy, and convergence failures. Recently, reinforcement-learning-based sequential variable-node (VN) scheduling (RL-S) was shown to improve BP decoding by learning state-dependent update orders. However, the VN-by-VN nature of that approach offers limited within-iteration parallelism, since only one VN is updated at a time. In this paper, we propose a cluster-based extension of RL-S for QLDPC codes. The VNs are partitioned into fixed clusters, and at each scheduling step the RL agent selects one cluster to update, after which all VNs in that cluster are updated in parallel using the same pre-update incoming messages. To keep the tabular state space practical for large cluster sizes, we introduce a permutation-invariant cluster state based on a normalized histogram of local mismatch weights, followed by quantization. This representation makes the number of cluster states depend on the quantization resolution rather than the cluster size. We also develop the corresponding cluster-level Markov decision process, reward function, and Q-learning update. Numerical results on representative QLDPC codes show that the proposed clustered learned scheduling preserves most of the error-rate benefit of VN-level learned sequential scheduling while substantially reducing the number of scheduling decisions per BP iteration, thereby providing an attractive latency-parallelism tradeoff. △ Less

Submitted 22 July, 2026; originally announced July 2026.

Constituency Optimisation Through Hamiltonian Representation Of Mandates (COTHROM): Algorithmic Redistricting of Irish Election Boundaries

Authors: Ruaidhrí Campion, Matthew Fenlon, Joshua Cooney Mercedal, Casey Farren-Colloty, Eliza Somerville, Michael A. J. Mitchell

Abstract: Electoral redistricting in Ireland's Proportional Representation Single Transferable Vote (PR-STV) system faces the challenge of selecting an optimally representative set of electoral boundaries from an enormous set of possible configurations, and where ``representative'' is a delicate balance of constitutional objectives that are often in tension with one another. We present the first computation… ▽ More Electoral redistricting in Ireland's Proportional Representation Single Transferable Vote (PR-STV) system faces the challenge of selecting an optimally representative set of electoral boundaries from an enormous set of possible configurations, and where ``representative'' is a delicate balance of constitutional objectives that are often in tension with one another. We present the first computational framework for Irish electoral redistricting that systematically optimises across multiple constitutional requirements while making trade-offs explicit and quantifiable. The electoral redistricting problem is parsed using statistical physics, where constitutional objectives are considered as terms in a Potts Hamiltonian. Markov Chain Monte Carlo (MCMC) methods and simulated annealing are employed to minimise this objective function, systematically exploring this configuration space, with coupling constants as proxies for objective weightings. Multi Criterion Decision Analysis (MCDA) and Pareto Optimality is then utilised to remedy the ambiguity in choosing a certain objective weighting combination over others. With respect to proportional representation and compactness objectives evaluated in County Cork, COTHROM consistently improves on the existing legal constituency boundaries for a range of objective weightings. △ Less

Submitted 2 June, 2026; originally announced June 2026.

Comments: 35 pages, 16 figures

Dynamic Parameter Scheduling in Soft-Hard BPGD for Lossy Source Coding

Authors: Masoumeh Alinia, David G. M. Mitchell

Abstract: We investigate lossy source coding based on a soft-decision belief propagation guided decimation (BPGD) encoder for low-density generator matrix (LDGM) codes, referred to as \emph{soft-hard BPGD}. The performance of this encoder is highly sensitive to the choice of ``softness'' parameters, typically denoted by $(β,μ)$, which are conventionally tuned via exhaustive empirical sweeps. To reduce this… ▽ More We investigate lossy source coding based on a soft-decision belief propagation guided decimation (BPGD) encoder for low-density generator matrix (LDGM) codes, referred to as \emph{soft-hard BPGD}. The performance of this encoder is highly sensitive to the choice of ``softness'' parameters, typically denoted by $(β,μ)$, which are conventionally tuned via exhaustive empirical sweeps. To reduce this burden and to better align the algorithm with the evolving graphical structure during decimation, we introduce a \emph{dynamic scheduling} framework in which $(β,μ)$ are not fixed globally but change as decimation progresses. The schedule starts in a softer regime to encourage exploration and gradually hardens toward the end to promote convergence, similar to simulated annealing. We consider linear and exponential schedules, discuss their physical interpretation via an effective temperature viewpoint, and explain how they integrate with soft-hard BPGD without changing the order of magnitude of its complexity. Numerical experiments with irregular and semi-regular LDGM ensembles indicate improved rate-distortion performance and reduced non-convergence compared to constant-parameter baselines, while largely eliminating expensive grid searches for a single best pair $(β,μ)$. △ Less

Submitted 18 April, 2026; originally announced April 2026.

Learning to Decode Quantum LDPC Codes Via Belief Propagation

Authors: Mohsen Moradi, Vahid Nourozi, Salman Habib, David G. M. Mitchell

Abstract: Belief-propagation (BP) decoding for quantum low-density parity-check (QLDPC) codes is appealing due to its low complexity, yet it often exhibits convergence issues due to quantum degeneracy and short cycles that exist in the Tanner graph. To overcome this challenge, this paper proposes a reinforcement-learning (RL) approach that learns (offline) how to decode QLDPC codes based on sequential decod… ▽ More Belief-propagation (BP) decoding for quantum low-density parity-check (QLDPC) codes is appealing due to its low complexity, yet it often exhibits convergence issues due to quantum degeneracy and short cycles that exist in the Tanner graph. To overcome this challenge, this paper proposes a reinforcement-learning (RL) approach that learns (offline) how to decode QLDPC codes based on sequential decoding trajectories. The decoding is formulated as a Markov decision process with a local, syndrome-driven state representation of the underlying RL agent. To enable fast inference, critical for practical implementation, we incrementally update our RL-based QLDPC decoder using second-order neighborhoods that avoid global rescans. Simulation results on representative QLDPC codes demonstrate the superiority of the proposed RL-based QLDPC decoders in terms of performance and convergence speed when compared to flooding and random sequential schedules, while achieving performance competitive with state-of-the-art BP-based decoders at comparable complexity. △ Less

Submitted 10 March, 2026; originally announced March 2026.

Sequential BP-based Decoding of QLDPC Codes

Authors: Mohsen Moradi, Salman Habib, Vahid Nourozi, David G. M. Mitchell

Abstract: Quantum low-density parity-check (QLDPC) codes are a leading approach to quantum error correction, yet conventional belief propagation (BP) decoders often perform poorly, primarily due to non-convergence exacerbated by stabilizer constraints, which induce short cycles and degeneracy. We propose two scheduling variants, sequential check node scheduling (SCNS) and sequential variable node scheduling… ▽ More Quantum low-density parity-check (QLDPC) codes are a leading approach to quantum error correction, yet conventional belief propagation (BP) decoders often perform poorly, primarily due to non-convergence exacerbated by stabilizer constraints, which induce short cycles and degeneracy. We propose two scheduling variants, sequential check node scheduling (SCNS) and sequential variable node scheduling (SVNS), that improve BP's error-correction ability by processing check nodes (CNs) or variable nodes (VNs), respectively, in a fixed order, stabilizing message updates and reducing stalls. We also employ this technique to an improved BP-variant called BP guided decimation (BPGD), where symbols are progressively fixed during decoding iterations. Here, we demonstrate that the sequential BPGD (SBPGD) decoder can further improve the convergence properties and performance of the decoder. On standard QLDPC benchmarks under a Pauli-X noise model, our sequential schedules are shown to lower the block error rate relative to conventional BP, and SBPGD outperforms BPGD while using significantly fewer decimation rounds, translating to lower computational cost. These results demonstrate that changing the update schedule, without altering the code, can improve both the reliability and efficiency of BP-based decoding for QLDPC codes. For the [[1922,50,16]] C2 hypergraph-product code with independent X errors, SVNS-BP surpasses BP-OSD-0 in error correction at roughly the same complexity as standard BP. △ Less

Submitted 13 February, 2026; originally announced February 2026.

The Number of Cycles of Bi-regular Tanner Graphs in Terms of the Eigenvalues of the Adjacency Matrix

Authors: Roxana Smarandache, David G. M. Mitchell

Abstract: In this paper, we explore new connections between the cycles in the graph of low-density parity-check (LDPC) codes and the eigenvalues of the corresponding adjacency matrix. The resulting observations are used to derive fast, simple, recursive formulas for the number of cycles $N_{2k}$ of length $2k$, $k<g$, in a bi-regular graph of girth $g$. Moreover, we derive explicit formulas for $N_{2k}$,… ▽ More In this paper, we explore new connections between the cycles in the graph of low-density parity-check (LDPC) codes and the eigenvalues of the corresponding adjacency matrix. The resulting observations are used to derive fast, simple, recursive formulas for the number of cycles $N_{2k}$ of length $2k$, $k<g$, in a bi-regular graph of girth $g$. Moreover, we derive explicit formulas for $N_{2k}$, $k\leq 7$, in terms of the nonzero eigenvalues of the adjacency matrix. Throughout, we focus on the practically interesting class of bi-regular quasi-cyclic LDPC (QC-LDPC) codes, for which the eigenvalues can be obtained efficiently by applying techniques used for block-circulant matrices. △ Less

Submitted 8 January, 2026; originally announced January 2026.

Comments: 25 pages, submitted to IT Transactions

Do AI Models Perform Human-like Abstract Reasoning Across Modalities?

Authors: Claas Beger, Ryan Yi, Shuhao Fu, Kaleda Denton, Arseny Moskvichev, Sarah W. Tsai, Sivasankaran Rajamanickam, Melanie Mitchell

Abstract: OpenAI's o3-preview reasoning model exceeded human accuracy on the ARC-AGI-1 benchmark, but does that mean state-of-the-art models recognize and reason with the abstractions the benchmark was designed to test? Here we investigate abstraction abilities of AI models using the closely related but simpler ConceptARC benchmark. Our evaluations vary input modality (textual vs. visual), use of external P… ▽ More OpenAI's o3-preview reasoning model exceeded human accuracy on the ARC-AGI-1 benchmark, but does that mean state-of-the-art models recognize and reason with the abstractions the benchmark was designed to test? Here we investigate abstraction abilities of AI models using the closely related but simpler ConceptARC benchmark. Our evaluations vary input modality (textual vs. visual), use of external Python tools, and reasoning effort. Beyond output accuracy, we evaluate the natural-language rules that models generate to explain their solutions, enabling us to assess whether models recognize the abstractions that ConceptARC was designed to elicit. We show that the best models' rules are frequently based on surface-level ``shortcuts,'' capturing intended abstractions considerably less often than humans. In the visual modality, AI models' output accuracy drops sharply; however, our rule-level analysis reveals that a substantial share of their rules capture the intended abstractions, even as the models struggle to apply these concepts to generate correct solutions. In short, we show that using accuracy alone to evaluate abstract reasoning can substantially overestimate AI capabilities in textual modalities and underestimate it in visual modalities. Our results offer a more faithful picture of AI models' abstract reasoning abilities and a more principled way to track progress toward human-like, abstraction-centered intelligence. △ Less

Submitted 2 February, 2026; v1 submitted 2 October, 2025; originally announced October 2025.

Comments: 9 pages, 3 figures

Generalized Quasi-Cyclic LDPC Codes: Design and Efficient Encoding

Authors: Roxana Smarandache, David G. M. Mitchell, Anthony Gómez-Fonseca

Abstract: Generalized low-density parity-check (GLDPC) codes, where single parity-check constraints on the code bits are replaced with generalized constraints (an arbitrary linear code), are a promising class of codes for low-latency communication. The block error rate performance of the GLDPC codes, combined with a complementary outer code, has been shown to outperform a variety of state-of-the-art code an… ▽ More Generalized low-density parity-check (GLDPC) codes, where single parity-check constraints on the code bits are replaced with generalized constraints (an arbitrary linear code), are a promising class of codes for low-latency communication. The block error rate performance of the GLDPC codes, combined with a complementary outer code, has been shown to outperform a variety of state-of-the-art code and decoder designs with suitable lengths and rates for the 5G ultra-reliable low-latency communication (URLLC) regime. A major drawback of these codes is that it is not known how to construct appropriate polynomial matrices to encode them efficiently. In this paper, we analyze practical constructions of quasi-cyclic GLDPC (QC-GLDPC) codes and show how to construct polynomial generator matrices in various forms using minors of the polynomial matrix. The approach can be applied to fully generalized matrices or partially generalized (with mixed constraint node types) to find better performance/rate trade-offs. The resulting encoding matrices are presented in useful forms that facilitate efficient implementation. The rich substructure displayed also provides us with new methods of determining low weight codewords, providing lower and upper bounds on the minimum distance and often giving those of weight equal to the minimum distance. Based on the minors of the polynomial parity-check matrix, we also give a formula for the rank of any parity-check matrix representing a QC-LDPC or QC-GLDPC code, and hence, the dimension of the code. Finally, we show that by applying double graph-liftings, the code parameters can be improved without affecting the ability to obtain a polynomial generator matrix. △ Less

Submitted 9 August, 2025; originally announced August 2025.

Rigor in AI: Doing Rigorous AI Work Requires a Broader, Responsible AI-Informed Conception of Rigor

Authors: Alexandra Olteanu, Su Lin Blodgett, Agathe Balayn, Angelina Wang, Fernando Diaz, Flavio du Pin Calmon, Margaret Mitchell, Michael Ekstrand, Reuben Binns, Solon Barocas

Abstract: In AI research and practice, rigor remains largely understood in terms of methodological rigor -- such as whether mathematical, statistical, or computational methods are correctly applied. We argue that this narrow conception of rigor has contributed to the concerns raised by the responsible AI community, including overblown claims about the capabilities of AI systems. Our position is that a broad… ▽ More In AI research and practice, rigor remains largely understood in terms of methodological rigor -- such as whether mathematical, statistical, or computational methods are correctly applied. We argue that this narrow conception of rigor has contributed to the concerns raised by the responsible AI community, including overblown claims about the capabilities of AI systems. Our position is that a broader conception of what rigorous AI research and practice should entail is needed. We believe such a conception -- in addition to a more expansive understanding of (1) methodological rigor -- should include aspects related to (2) what background knowledge informs what to work on (epistemic rigor); (3) how disciplinary, community, or personal norms, standards, or beliefs influence the work (normative rigor); (4) how clearly articulated the theoretical constructs under use are (conceptual rigor); (5) what is reported and how (reporting rigor); and (6) how well-supported the inferences from existing evidence are (interpretative rigor). In doing so, we also provide useful language and a framework for much-needed dialogue about the AI community's work by researchers, policymakers, journalists, and other stakeholders. △ Less

Submitted 25 November, 2025; v1 submitted 17 June, 2025; originally announced June 2025.

Comments: 21 pages, 1 figure, 1 table, accepted at NeurIPS'25 position papers track

Large Language Models and Emergence: A Complex Systems Perspective

Authors: David C. Krakauer, John W. Krakauer, Melanie Mitchell

Abstract: Emergence is a concept in complexity science that describes how many-body systems manifest novel higher-level properties, properties that can be described by replacing high-dimensional mechanisms with lower-dimensional effective variables and theories. This is captured by the idea "more is different". Intelligence is a consummate emergent property manifesting increasingly efficient -- cheaper and… ▽ More Emergence is a concept in complexity science that describes how many-body systems manifest novel higher-level properties, properties that can be described by replacing high-dimensional mechanisms with lower-dimensional effective variables and theories. This is captured by the idea "more is different". Intelligence is a consummate emergent property manifesting increasingly efficient -- cheaper and faster -- uses of emergent capabilities to solve problems. This is captured by the idea "less is more". In this paper, we first examine claims that Large Language Models exhibit emergent capabilities, reviewing several approaches to quantifying emergence, and secondly ask whether LLMs possess emergent intelligence. △ Less

Submitted 10 June, 2025; originally announced June 2025.

Toward an Evaluation Science for Generative AI Systems

Authors: Laura Weidinger, Inioluwa Deborah Raji, Hanna Wallach, Margaret Mitchell, Angelina Wang, Olawale Salaudeen, Rishi Bommasani, Deep Ganguli, Sanmi Koyejo, William Isaac

Abstract: There is an increasing imperative to anticipate and understand the performance and safety of generative AI systems in real-world deployment contexts. However, the current evaluation ecosystem is insufficient: Commonly used static benchmarks face validity challenges, and ad hoc case-by-case audits rarely scale. In this piece, we advocate for maturing an evaluation science for generative AI systems.… ▽ More There is an increasing imperative to anticipate and understand the performance and safety of generative AI systems in real-world deployment contexts. However, the current evaluation ecosystem is insufficient: Commonly used static benchmarks face validity challenges, and ad hoc case-by-case audits rarely scale. In this piece, we advocate for maturing an evaluation science for generative AI systems. While generative AI creates unique challenges for system safety engineering and measurement science, the field can draw valuable insights from the development of safety evaluation practices in other fields, including transportation, aerospace, and pharmaceutical engineering. In particular, we present three key lessons: Evaluation metrics must be applicable to real-world performance, metrics must be iteratively refined, and evaluation institutions and norms must be established. Applying these insights, we outline a concrete path toward a more rigorous approach for evaluating generative AI systems. △ Less

Submitted 12 March, 2025; v1 submitted 7 March, 2025; originally announced March 2025.

Comments: First two authors contributed equally to this work

High-Rate Spatially Coupled LDPC Codes Based on Massey's Convolutional Self-Orthogonal Codes

Authors: Daniel J. Costello, Jr., Min Zhu, David G. M. Mitchell, Michael Lentmaier

Abstract: In this paper, we study a new class of high-rate spatially coupled LDPC (SC-LDPC) codes based on the convolutional self-orthogonal codes (CSOCs) first introduced by Massey. The SC-LDPC codes are constructed by treating the irregular graph corresponding to the parity-check matrix of a systematic rate R = (n - 1)/n CSOC as a convolutional protograph. The protograph can then be lifted using permutati… ▽ More In this paper, we study a new class of high-rate spatially coupled LDPC (SC-LDPC) codes based on the convolutional self-orthogonal codes (CSOCs) first introduced by Massey. The SC-LDPC codes are constructed by treating the irregular graph corresponding to the parity-check matrix of a systematic rate R = (n - 1)/n CSOC as a convolutional protograph. The protograph can then be lifted using permutation matrices to generate a high-rate SC-LDPC code whose strength depends on the lifting factor. The SC-LDPC codes constructed in this fashion can be decoded using iterative belief propagation (BP) based sliding window decoding (SWD). A non-systematic version of a CSOC parity-check matrix is then proposed by making a slight modification to the systematic construction. The non-systematic parity-check matrix corresponds to a regular protograph whose degree profile depends on the rate and error-correcting capability of the underlying CSOC. Even though the parity-check matrix is in non-systematic form, we show how systematic encoding can still be performed. We also show that the non-systematic convolutional protograph has a guaranteed girth and free distance and that these properties carry over to the lifted versions. Finally, numerical results are included demonstrating that CSOC-based SC-LDPC codes (i) achieve excellent performance at very high rates, (ii) have performance at least as good as that of SC-LDPC codes constructed from convolutional protographs commonly found in the literature, and (iii) have iterative decoding thresholds comparable to those of existing SC-LDPC code designs. △ Less

Submitted 17 February, 2025; v1 submitted 5 February, 2025; originally announced February 2025.

Stop treating `AGI' as the north-star goal of AI research

Authors: Borhane Blili-Hamelin, Christopher Graziul, Leif Hancox-Li, Hananel Hazan, El-Mahdi El-Mhamdi, Avijit Ghosh, Katherine Heller, Jacob Metcalf, Fabricio Murai, Eryk Salvaggio, Andrew Smart, Todd Snider, Mariame Tighanimine, Talia Ringer, Margaret Mitchell, Shiri Dori-Hacohen

Abstract: The AI research community plays a vital role in shaping the scientific, engineering, and societal goals of AI research. In this position paper, we argue that focusing on the highly contested topic of `artificial general intelligence' (`AGI') undermines our ability to choose effective goals. We identify six key traps -- obstacles to productive goal setting -- that are aggravated by AGI discourse: I… ▽ More The AI research community plays a vital role in shaping the scientific, engineering, and societal goals of AI research. In this position paper, we argue that focusing on the highly contested topic of `artificial general intelligence' (`AGI') undermines our ability to choose effective goals. We identify six key traps -- obstacles to productive goal setting -- that are aggravated by AGI discourse: Illusion of Consensus, Supercharging Bad Science, Presuming Value-Neutrality, Goal Lottery, Generality Debt, and Normalized Exclusion. To avoid these traps, we argue that the AI research community needs to (1) prioritize specificity in engineering and societal goals, (2) center pluralism about multiple worthwhile approaches to multiple valuable goals, and (3) foster innovation through greater inclusion of disciplines and communities. Therefore, the AI research community needs to stop treating `AGI' as the north-star goal of AI research. △ Less

Submitted 7 July, 2025; v1 submitted 5 February, 2025; originally announced February 2025.

Comments: Position Paper accepted to ICML 2025. OpenReview: https://openreview.net/forum?id=1RlrtH6ydW

Fully Autonomous AI Agents Should Not be Developed

Authors: Margaret Mitchell, Avijit Ghosh, Alexandra Sasha Luccioni, Giada Pistilli

Abstract: This paper argues that fully autonomous AI agents should not be developed. In support of this position, we build from prior scientific literature and current product marketing to delineate different AI agent levels and detail the ethical values at play in each, documenting trade-offs in potential benefits and risks. Our analysis reveals that risks to people increase with the autonomy of a system:… ▽ More This paper argues that fully autonomous AI agents should not be developed. In support of this position, we build from prior scientific literature and current product marketing to delineate different AI agent levels and detail the ethical values at play in each, documenting trade-offs in potential benefits and risks. Our analysis reveals that risks to people increase with the autonomy of a system: The more control a user cedes to an AI agent, the more risks to people arise. Particularly concerning are safety risks, which affect human life and impact further values. △ Less

Submitted 19 October, 2025; v1 submitted 4 February, 2025; originally announced February 2025.

Evaluating the Robustness of Analogical Reasoning in Large Language Models

Authors: Martha Lewis, Melanie Mitchell

Abstract: LLMs have performed well on several reasoning benchmarks, including ones that test analogical reasoning abilities. However, there is debate on the extent to which they are performing general abstract reasoning versus employing non-robust processes, e.g., that overly rely on similarity to pre-training data. Here we investigate the robustness of analogy-making abilities previously claimed for LLMs o… ▽ More LLMs have performed well on several reasoning benchmarks, including ones that test analogical reasoning abilities. However, there is debate on the extent to which they are performing general abstract reasoning versus employing non-robust processes, e.g., that overly rely on similarity to pre-training data. Here we investigate the robustness of analogy-making abilities previously claimed for LLMs on three of four domains studied by Webb, Holyoak, and Lu (2023): letter-string analogies, digit matrices, and story analogies. For each domain we test humans and GPT models on robustness to variants of the original analogy problems that test the same abstract reasoning abilities but are likely dissimilar from tasks in the pre-training data. The performance of a system that uses robust abstract reasoning should not decline substantially on these variants. On simple letter-string analogies, we find that while the performance of humans remains high for two types of variants we tested, the GPT models' performance declines sharply. This pattern is less pronounced as the complexity of these problems is increased, as both humans and GPT models perform poorly on both the original and variant problems requiring more complex analogies. On digit-matrix problems, we find a similar pattern but only on one out of the two types of variants we tested. On story-based analogy problems, we find that, unlike humans, the performance of GPT models are susceptible to answer-order effects, and that GPT models also may be more sensitive than humans to paraphrasing. This work provides evidence that LLMs often lack the robustness of zero-shot human analogy-making, exhibiting brittleness on most of the variations we tested. More generally, this work points to the importance of carefully evaluating AI systems not only for accuracy but also robustness when testing their cognitive capabilities. △ Less

Submitted 21 November, 2024; originally announced November 2024.

Comments: 31 pages, 13 figures. arXiv admin note: text overlap with arXiv:2402.08955

Traceable random numbers from a nonlocal quantum advantage

Authors: Gautam A. Kavuri, Jasper Palfree, Dileep V. Reddy, Yanbao Zhang, Joshua C. Bienfang, Michael D. Mazurek, Mohammad A. Alhejji, Aliza U. Siddiqui, Joseph M. Cavanagh, Aagam Dalal, Carlos Abellán, Waldimar Amaya, Morgan W. Mitchell, Katherine E. Stange, Paul D. Beale, Luís T. A. N. Brandão, Harold Booth, René Peralta, Sae Woo Nam, Richard P. Mirin, Martin J. Stevens, Emanuel Knill, Lynden K. Shalm

Abstract: The unpredictability of random numbers is fundamental to both digital security and applications that fairly distribute resources. However, existing random number generators have limitations-the generation processes cannot be fully traced, audited, and certified to be unpredictable. The algorithmic steps used in pseudorandom number generators are auditable, but they cannot guarantee that their outp… ▽ More The unpredictability of random numbers is fundamental to both digital security and applications that fairly distribute resources. However, existing random number generators have limitations-the generation processes cannot be fully traced, audited, and certified to be unpredictable. The algorithmic steps used in pseudorandom number generators are auditable, but they cannot guarantee that their outputs were a priori unpredictable given knowledge of the initial seed. Device-independent quantum random number generators can ensure that the source of randomness was unknown beforehand, but the steps used to extract the randomness are vulnerable to tampering. Here, for the first time, we demonstrate a fully traceable random number generation protocol based on device-independent techniques. Our protocol extracts randomness from unpredictable non-local quantum correlations, and uses distributed intertwined hash chains to cryptographically trace and verify the extraction process. This protocol is at the heart of a public traceable and certifiable quantum randomness beacon that we have launched. Over the first 40 days of operation, we completed the protocol 7434 out of 7454 attempts -- a success rate of 99.7%. Each time the protocol succeeded, the beacon emitted a pulse of 512 bits of traceable randomness. The bits are certified to be uniform with error times actual success probability bounded by $2^{-64}$. The generation of certifiable and traceable randomness represents one of the first public services that operates with an entanglement-derived advantage over comparable classical approaches. △ Less

Submitted 7 November, 2024; originally announced November 2024.

Comments: 40 pages, 4 main figures, 10 supplementary figures

Imagining and building wise machines: The centrality of AI metacognition

Authors: Samuel G. B. Johnson, Amir-Hossein Karimi, Yoshua Bengio, Nick Chater, Tobias Gerstenberg, Kate Larson, Sydney Levine, Melanie Mitchell, Iyad Rahwan, Bernhard Schölkopf, Igor Grossmann

Abstract: Although AI has become increasingly smart, its wisdom has not kept pace. In this article, we examine what is known about human wisdom and sketch a vision of its AI counterpart. We analyze human wisdom as a set of strategies for solving intractable problems-those outside the scope of analytic techniques-including both object-level strategies like heuristics [for managing problems] and metacognitive… ▽ More Although AI has become increasingly smart, its wisdom has not kept pace. In this article, we examine what is known about human wisdom and sketch a vision of its AI counterpart. We analyze human wisdom as a set of strategies for solving intractable problems-those outside the scope of analytic techniques-including both object-level strategies like heuristics [for managing problems] and metacognitive strategies like intellectual humility, perspective-taking, or context-adaptability [for managing object-level strategies]. We argue that AI systems particularly struggle with metacognition; improved metacognition would lead to AI more robust to novel environments, explainable to users, cooperative with others, and safer in risking fewer misaligned goals with human users. We discuss how wise AI might be benchmarked, trained, and implemented. △ Less

Submitted 7 January, 2026; v1 submitted 4 November, 2024; originally announced November 2024.

Comments: 23 pages, 2 figures, 2 tables

Can Large Language Models generalize analogy solving like children can?

Authors: Claire E. Stevenson, Alexandra Pafford, Han L. J. van der Maas, Melanie Mitchell

Abstract: In people, the ability to solve analogies such as "body : feet :: table : ?" emerges in childhood, and appears to transfer easily to other domains, such as the visual domain "( : ) :: < : ?". Recent research shows that large language models (LLMs) can solve various forms of analogies. However, can LLMs generalize analogy solving to new domains like people can? To investigate this, we had children,… ▽ More In people, the ability to solve analogies such as "body : feet :: table : ?" emerges in childhood, and appears to transfer easily to other domains, such as the visual domain "( : ) :: < : ?". Recent research shows that large language models (LLMs) can solve various forms of analogies. However, can LLMs generalize analogy solving to new domains like people can? To investigate this, we had children, adults, and LLMs solve a series of letter-string analogies (e.g., a b : a c :: j k : ?) in the Latin alphabet, in a near transfer domain (Greek alphabet), and a far transfer domain (list of symbols). Children and adults easily generalized their knowledge to unfamiliar domains, whereas LLMs did not. This key difference between human and AI performance is evidence that these LLMs still struggle with robust human-like analogical transfer. △ Less

Submitted 6 October, 2025; v1 submitted 4 November, 2024; originally announced November 2024.

Comments: Accepted to Transactions of the Association for Computational Linguistics (TACL)

Spectrally Efficient LDPC Codes For IRIG-106 Waveforms via Random Puncturing

Authors: Andrew D. Cummins, David G. M. Mitchell, Erik Perrins

Abstract: Low-density parity-check (LDPC) codes form part of the IRIG-106 standard and have been successfully deployed for the Telemetry Group version of shaped-offset quadrature phase shift keying (SOQPSK-TG) modulation. Recently, LDPC code solutions have been proposed and optimized for continuous phase modulations (CPMs), including the pulse code modulation/frequency modulation (PCM/FM) and the multi-h CP… ▽ More Low-density parity-check (LDPC) codes form part of the IRIG-106 standard and have been successfully deployed for the Telemetry Group version of shaped-offset quadrature phase shift keying (SOQPSK-TG) modulation. Recently, LDPC code solutions have been proposed and optimized for continuous phase modulations (CPMs), including the pulse code modulation/frequency modulation (PCM/FM) and the multi-h CPM developed by the Advanced Range TeleMetry program (ARTM CPM). These codes were shown to perform around one dB from the respective channel capacities of these modulations. In this paper, we consider the effect of random puncturing of these LDPC codes to further improve spectrum efficiency. We present numerical simulation results that affirm the robust decoding performance promised by LDPC codes designed for ARTM CPM. △ Less

Submitted 8 October, 2024; originally announced October 2024.

Comments: Accepted for inclusion in the 2024 International Telemetry Conference

Constructing the CORD-19 Vaccine Dataset

Authors: Manisha Singh, Divy Sharma, Alonso Ma, Bridget Tyree, Margaret Mitchell

Abstract: We introduce new dataset 'CORD-19-Vaccination' to cater to scientists specifically looking into COVID-19 vaccine-related research. This dataset is extracted from CORD-19 dataset [Wang et al., 2020] and augmented with new columns for language detail, author demography, keywords, and topic per paper. Facebook's fastText model is used to identify languages [Joulin et al., 2016]. To establish author d… ▽ More We introduce new dataset 'CORD-19-Vaccination' to cater to scientists specifically looking into COVID-19 vaccine-related research. This dataset is extracted from CORD-19 dataset [Wang et al., 2020] and augmented with new columns for language detail, author demography, keywords, and topic per paper. Facebook's fastText model is used to identify languages [Joulin et al., 2016]. To establish author demography (author affiliation, lab/institution location, and lab/institution country columns) we processed the JSON file for each paper and then further enhanced using Google's search API to determine country values. 'Yake' was used to extract keywords from the title, abstract, and body of each paper and the LDA (Latent Dirichlet Allocation) algorithm was used to add topic information [Campos et al., 2020, 2018a,b]. To evaluate the dataset, we demonstrate a question-answering task like the one used in the CORD-19 Kaggle challenge [Goldbloom et al., 2022]. For further evaluation, sequential sentence classification was performed on each paper's abstract using the model from Dernoncourt et al. [2016]. We partially hand annotated the training dataset and used a pre-trained BERT-PubMed layer. 'CORD- 19-Vaccination' contains 30k research papers and can be immensely valuable for NLP research such as text mining, information extraction, and question answering, specific to the domain of COVID-19 vaccine research. △ Less

Submitted 25 July, 2024; originally announced July 2024.

The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale

Authors: Guilherme Penedo, Hynek Kydlíček, Loubna Ben allal, Anton Lozhkov, Margaret Mitchell, Colin Raffel, Leandro Von Werra, Thomas Wolf

Abstract: The performance of a large language model (LLM) depends heavily on the quality and size of its pretraining dataset. However, the pretraining datasets for state-of-the-art open LLMs like Llama 3 and Mixtral are not publicly available and very little is known about how they were created. In this work, we introduce FineWeb, a 15-trillion token dataset derived from 96 Common Crawl snapshots that produ… ▽ More The performance of a large language model (LLM) depends heavily on the quality and size of its pretraining dataset. However, the pretraining datasets for state-of-the-art open LLMs like Llama 3 and Mixtral are not publicly available and very little is known about how they were created. In this work, we introduce FineWeb, a 15-trillion token dataset derived from 96 Common Crawl snapshots that produces better-performing LLMs than other open pretraining datasets. To advance the understanding of how best to curate high-quality pretraining datasets, we carefully document and ablate all of the design choices used in FineWeb, including in-depth investigations of deduplication and filtering strategies. In addition, we introduce FineWeb-Edu, a 1.3-trillion token collection of educational text filtered from FineWeb. LLMs pretrained on FineWeb-Edu exhibit dramatically better performance on knowledge- and reasoning-intensive benchmarks like MMLU and ARC. Along with our datasets, we publicly release our data curation codebase and all of the models trained during our ablation experiments. △ Less

Submitted 31 October, 2024; v1 submitted 25 June, 2024; originally announced June 2024.

CIVICS: Building a Dataset for Examining Culturally-Informed Values in Large Language Models

Authors: Giada Pistilli, Alina Leidinger, Yacine Jernite, Atoosa Kasirzadeh, Alexandra Sasha Luccioni, Margaret Mitchell

Abstract: This paper introduces the "CIVICS: Culturally-Informed & Values-Inclusive Corpus for Societal impacts" dataset, designed to evaluate the social and cultural variation of Large Language Models (LLMs) across multiple languages and value-sensitive topics. We create a hand-crafted, multilingual dataset of value-laden prompts which address specific socially sensitive topics, including LGBTQI rights, so… ▽ More This paper introduces the "CIVICS: Culturally-Informed & Values-Inclusive Corpus for Societal impacts" dataset, designed to evaluate the social and cultural variation of Large Language Models (LLMs) across multiple languages and value-sensitive topics. We create a hand-crafted, multilingual dataset of value-laden prompts which address specific socially sensitive topics, including LGBTQI rights, social welfare, immigration, disability rights, and surrogacy. CIVICS is designed to generate responses showing LLMs' encoded and implicit values. Through our dynamic annotation processes, tailored prompt design, and experiments, we investigate how open-weight LLMs respond to value-sensitive issues, exploring their behavior across diverse linguistic and cultural contexts. Using two experimental set-ups based on log-probabilities and long-form responses, we show social and cultural variability across different LLMs. Specifically, experiments involving long-form responses demonstrate that refusals are triggered disparately across models, but consistently and more frequently in English or translated statements. Moreover, specific topics and sources lead to more pronounced differences across model answers, particularly on immigration, LGBTQI rights, and social welfare. As shown by our experiments, the CIVICS dataset aims to serve as a tool for future research, promoting reproducibility and transparency across broader linguistic settings, and furthering the development of AI technologies that respect and reflect global cultural diversities and value pluralism. The CIVICS dataset and tools will be made available upon publication under open licenses; an anonymized version is currently available at https://huggingface.co/CIVICS-dataset. △ Less

Submitted 22 May, 2024; originally announced May 2024.

Using Counterfactual Tasks to Evaluate the Generality of Analogical Reasoning in Large Language Models

Authors: Martha Lewis, Melanie Mitchell

Abstract: Large language models (LLMs) have performed well on several reasoning benchmarks, including ones that test analogical reasoning abilities. However, it has been debated whether they are actually performing humanlike abstract reasoning or instead employing less general processes that rely on similarity to what has been seen in their training data. Here we investigate the generality of analogy-making… ▽ More Large language models (LLMs) have performed well on several reasoning benchmarks, including ones that test analogical reasoning abilities. However, it has been debated whether they are actually performing humanlike abstract reasoning or instead employing less general processes that rely on similarity to what has been seen in their training data. Here we investigate the generality of analogy-making abilities previously claimed for LLMs (Webb, Holyoak, & Lu, 2023). We take one set of analogy problems used to evaluate LLMs and create a set of "counterfactual" variants-versions that test the same abstract reasoning abilities but that are likely dissimilar from any pre-training data. We test humans and three GPT models on both the original and counterfactual problems, and show that, while the performance of humans remains high for all the problems, the GPT models' performance declines sharply on the counterfactual set. This work provides evidence that, despite previously reported successes of LLMs on analogical reasoning, these models lack the robustness and generality of human analogy-making. △ Less

Submitted 14 February, 2024; originally announced February 2024.

PAC Code Rate-Profile Design Using Search-Constrained Optimization Algorithms

Authors: Mohsen Moradi, David G. M. Mitchell

Abstract: In this paper, we introduce a novel rate-profile design based on search-constrained optimization techniques to assess the performance of polarization-adjusted convolutional (PAC) codes under Fano (sequential) decoding. The results demonstrate that the resulting PAC code offers much reduced computational complexity compared to a construction based on a conventional genetic algorithm without a perfo… ▽ More In this paper, we introduce a novel rate-profile design based on search-constrained optimization techniques to assess the performance of polarization-adjusted convolutional (PAC) codes under Fano (sequential) decoding. The results demonstrate that the resulting PAC code offers much reduced computational complexity compared to a construction based on a conventional genetic algorithm without a performance loss in error-correction performance. As the fitness function of our algorithm, we propose an adaptive successive cancellation list decoding algorithm to determine the weight distribution of the rate profiles. The simulation results indicate that, for a PAC(256, 128) code, only 8% of the population requires that their fitness function be evaluated with a large list size. This represents an improvement of almost 92% over a conventional evolutionary algorithm. For a PAC(64, 32) code, this improvement is about 99%. We also plotted the performance of the high-rate PAC(128, 105) and PAC(64, 51) codes, and the results show that they exhibit superior performance compared to other algorithms. △ Less

Submitted 18 January, 2024; originally announced January 2024.

MorpheusNet: Resource efficient sleep stage classifier for embedded on-line systems

Authors: Ali Kavoosi, Morgan P. Mitchell, Raveen Kariyawasam, John E. Fleming, Penny Lewis, Heidi Johansen-Berg, Hayriye Cagnan, Timothy Denison

Abstract: Sleep Stage Classification (SSC) is a labor-intensive task, requiring experts to examine hours of electrophysiological recordings for manual classification. This is a limiting factor when it comes to leveraging sleep stages for therapeutic purposes. With increasing affordability and expansion of wearable devices, automating SSC may enable deployment of sleep-based therapies at scale. Deep Learning… ▽ More Sleep Stage Classification (SSC) is a labor-intensive task, requiring experts to examine hours of electrophysiological recordings for manual classification. This is a limiting factor when it comes to leveraging sleep stages for therapeutic purposes. With increasing affordability and expansion of wearable devices, automating SSC may enable deployment of sleep-based therapies at scale. Deep Learning has gained increasing attention as a potential method to automate this process. Previous research has shown accuracy comparable to manual expert scores. However, previous approaches require sizable amount of memory and computational resources. This constrains the ability to classify in real time and deploy models on the edge. To address this gap, we aim to provide a model capable of predicting sleep stages in real-time, without requiring access to external computational sources (e.g., mobile phone, cloud). The algorithm is power efficient to enable use on embedded battery powered systems. Our compact sleep stage classifier can be deployed on most off-the-shelf microcontrollers (MCU) with constrained hardware settings. This is due to the memory footprint of our approach requiring significantly fewer operations. The model was tested on three publicly available data bases and achieved performance comparable to the state of the art, whilst reducing model complexity by orders of magnitude (up to 280 times smaller compared to state of the art). We further optimized the model with quantization of parameters to 8 bits with only an average drop of 0.95% in accuracy. When implemented in firmware, the quantized model achieves a latency of 1.6 seconds on an Arm CortexM4 processor, allowing its use for on-line SSC-based therapies. △ Less

Submitted 14 January, 2024; originally announced January 2024.

Comments: This paper was presented at the 2023 IEEE conference on Systems, Man, and Cybernetics (SMC)

Perspectives on the State and Future of Deep Learning - 2023

Authors: Micah Goldblum, Anima Anandkumar, Richard Baraniuk, Tom Goldstein, Kyunghyun Cho, Zachary C Lipton, Melanie Mitchell, Preetum Nakkiran, Max Welling, Andrew Gordon Wilson

Abstract: The goal of this series is to chronicle opinions and issues in the field of machine learning as they stand today and as they change over time. The plan is to host this survey periodically until the AI singularity paperclip-frenzy-driven doomsday, keeping an updated list of topical questions and interviewing new community members for each edition. In this issue, we probed people's opinions on inter… ▽ More The goal of this series is to chronicle opinions and issues in the field of machine learning as they stand today and as they change over time. The plan is to host this survey periodically until the AI singularity paperclip-frenzy-driven doomsday, keeping an updated list of topical questions and interviewing new community members for each edition. In this issue, we probed people's opinions on interpretable AI, the value of benchmarking in modern NLP, the state of progress towards understanding deep learning, and the future of academia. △ Less

Submitted 18 December, 2023; v1 submitted 7 December, 2023; originally announced December 2023.

120 GOPS Photonic Tensor Core in Thin-film Lithium Niobate for Inference and in-situ Training

Authors: Zhongjin Lin, Bhavin J. Shastri, Shangxuan Yu, Jingxiang Song, Yuntao Zhu, Arman Safarnejadian, Wangning Cai, Yanmei Lin, Wei Ke, Mustafa Hammood, Tianye Wang, Mengyue Xu, Zibo Zheng, Mohammed Al-Qadasi, Omid Esmaeeli, Mohamed Rahim, Grzegorz Pakulski, Jens Schmid, Pedro Barrios, Weihong Jiang, Hugh Morison, Matthew Mitchell, Xun Guan, Nicolas A. F. Jaeger, Leslie A. n Rusch , et al. (5 additional authors not shown)

Abstract: Photonics offers a transformative approach to artificial intelligence (AI) and neuromorphic computing by enabling low-latency, high-speed, and energy-efficient computations. However, conventional photonic tensor cores face significant challenges in constructing large-scale photonic neuromorphic networks. Here, we propose a fully integrated photonic tensor core, consisting of only two thin-film lit… ▽ More Photonics offers a transformative approach to artificial intelligence (AI) and neuromorphic computing by enabling low-latency, high-speed, and energy-efficient computations. However, conventional photonic tensor cores face significant challenges in constructing large-scale photonic neuromorphic networks. Here, we propose a fully integrated photonic tensor core, consisting of only two thin-film lithium niobate (TFLN) modulators, a III-V laser, and a charge-integration photoreceiver. Despite its simple architecture, it is capable of implementing an entire layer of a neural network with a computational speed of 120 GOPS, while also allowing flexible adjustment of the number of inputs (fan-in) and outputs (fan-out). Our tensor core supports rapid in-situ training with a weight update speed of 60 GHz. Furthermore, it successfully classifies (supervised learning) and clusters (unsupervised learning) 112 * 112-pixel images through in-situ training. To enable in-situ training for clustering AI tasks, we offer a solution for performing multiplications between two negative numbers. △ Less

Submitted 8 October, 2024; v1 submitted 28 November, 2023; originally announced November 2023.

Comments: 21 pages, 6 figures

MSC Class: 78A05

Comparing Humans, GPT-4, and GPT-4V On Abstraction and Reasoning Tasks

Authors: Melanie Mitchell, Alessandro B. Palmarini, Arseny Moskvichev

Abstract: We explore the abstract reasoning abilities of text-only and multimodal versions of GPT-4, using the ConceptARC benchmark [10], which is designed to evaluate robust understanding and reasoning with core-knowledge concepts. We extend the work of Moskvichev et al. [10] by evaluating GPT-4 on more detailed, one-shot prompting (rather than simple, zero-shot prompts) with text versions of ConceptARC ta… ▽ More We explore the abstract reasoning abilities of text-only and multimodal versions of GPT-4, using the ConceptARC benchmark [10], which is designed to evaluate robust understanding and reasoning with core-knowledge concepts. We extend the work of Moskvichev et al. [10] by evaluating GPT-4 on more detailed, one-shot prompting (rather than simple, zero-shot prompts) with text versions of ConceptARC tasks, and by evaluating GPT-4V, the multimodal version of GPT-4, on zero- and one-shot prompts using image versions of the simplest tasks. Our experimental results support the conclusion that neither version of GPT-4 has developed robust abstraction abilities at humanlike levels. △ Less

Submitted 11 December, 2023; v1 submitted 13 November, 2023; originally announced November 2023.

Comments: Corrected Figure 3 (extra spaces were replaced by commas, which were lost in original formatting)

Journal ref: Proceedings of the LLM-CP Workshop, AAAI 2024

SmartPlay: A Benchmark for LLMs as Intelligent Agents

Authors: Yue Wu, Xuan Tang, Tom M. Mitchell, Yuanzhi Li

Abstract: Recent large language models (LLMs) have demonstrated great potential toward intelligent agents and next-gen automation, but there currently lacks a systematic benchmark for evaluating LLMs' abilities as agents. We introduce SmartPlay: both a challenging benchmark and a methodology for evaluating LLMs as agents. SmartPlay consists of 6 different games, including Rock-Paper-Scissors, Tower of Hanoi… ▽ More Recent large language models (LLMs) have demonstrated great potential toward intelligent agents and next-gen automation, but there currently lacks a systematic benchmark for evaluating LLMs' abilities as agents. We introduce SmartPlay: both a challenging benchmark and a methodology for evaluating LLMs as agents. SmartPlay consists of 6 different games, including Rock-Paper-Scissors, Tower of Hanoi, Minecraft. Each game features a unique setting, providing up to 20 evaluation settings and infinite environment variations. Each game in SmartPlay uniquely challenges a subset of 9 important capabilities of an intelligent LLM agent, including reasoning with object dependencies, planning ahead, spatial reasoning, learning from history, and understanding randomness. The distinction between the set of capabilities each game test allows us to analyze each capability separately. SmartPlay serves not only as a rigorous testing ground for evaluating the overall performance of LLM agents but also as a road-map for identifying gaps in current methodologies. We release our benchmark at github.com/Microsoft/SmartPlay △ Less

Submitted 17 March, 2024; v1 submitted 2 October, 2023; originally announced October 2023.

Soft Air Pocket Force Sensors for Large Scale Flexible Robots

Authors: Michael R. Mitchell, Ciera McFarland, Margaret M. Coad

Abstract: Flexible robots have advantages over rigid robots in their ability to conform physically to their environment and to form a wide variety of shapes. Sensing the force applied by or to flexible robots is useful for both navigation and manipulation tasks, but it is challenging due to the need for the sensors to withstand the robots' shape change without encumbering their functionality. Also, for robo… ▽ More Flexible robots have advantages over rigid robots in their ability to conform physically to their environment and to form a wide variety of shapes. Sensing the force applied by or to flexible robots is useful for both navigation and manipulation tasks, but it is challenging due to the need for the sensors to withstand the robots' shape change without encumbering their functionality. Also, for robots with long or large bodies, the number of sensors required to cover the entire surface area of the robot body can be prohibitive due to high cost and complexity. We present a novel soft air pocket force sensor that is highly flexible, lightweight, relatively inexpensive, and easily scalable to various sizes. Our sensor produces a change in internal pressure that is linear with the applied force. We present results of experimental testing of how uncontrollable factors (contact location and contact area) and controllable factors (initial internal pressure, thickness, size, and number of interior seals) affect the sensitivity. We demonstrate our sensor applied to a vine robot-a soft inflatable robot that "grows" from the tip via eversion-and we show that the robot can successfully grow and steer towards an object with which it senses contact. △ Less

Submitted 26 July, 2023; originally announced July 2023.

Comments: M. R. Mitchell, C. McFarland, and M. M. Coad, "Soft Air Pocket Force Sensors for Large Scale Flexible Robots," in IEEE International Conference on Soft Robotics, 2023, pp. 1-8. Video: https://youtu.be/2De0htilW74

Reinforcement Learning for Sequential Decoding of Generalized LDPC Codes

Authors: Salman Habib, David G. M. Mitchell

Abstract: In this work, we propose reinforcement learning (RL) for sequential decoding of moderate length generalized low-density parity-check (GLDPC) codes. Here, sequential decoding refers to scheduling all the generalized constraint nodes (GCNs) and single parity-check nodes (SPCNs) of a GLDPC code serially in each iteration. A GLDPC decoding environment is modeled as a finite Markov decision process (MD… ▽ More In this work, we propose reinforcement learning (RL) for sequential decoding of moderate length generalized low-density parity-check (GLDPC) codes. Here, sequential decoding refers to scheduling all the generalized constraint nodes (GCNs) and single parity-check nodes (SPCNs) of a GLDPC code serially in each iteration. A GLDPC decoding environment is modeled as a finite Markov decision process (MDP) in which the state-space comprises of all possible sequences of hard-decision values of the variables nodes (VNs) connected to the scheduled GCN or SPCN, and the action-space of the MDP consists of all possible actions (GCN and SPCN scheduling). The goal of RL is to determine an optimized scheduling policy, i.e., one that results in a decoded codeword by minimizing the complexity of the belief propagation (BP) decoder. For training, we consider the proportion of correct bits at the output of the GCN or SPCN as a reward once it is scheduled. The expected rewards for scheduling all the GCNs/SPCNs in the code's Tanner graph are earned via BP decoding during the RL phase. The proposed RL-based decoding scheme is shown to significantly outperform the standard BP flooding decoder, as well as a sequential decoder in which the GCNs/SPCNs are scheduled randomly. △ Less

Submitted 25 July, 2023; originally announced July 2023.

Comments: accepted for publication at ISTC 2023. arXiv admin note: text overlap with arXiv:2112.13934

Evaluating the Social Impact of Generative AI Systems in Systems and Society

Authors: Irene Solaiman, Zeerak Talat, William Agnew, Lama Ahmad, Dylan Baker, Su Lin Blodgett, Canyu Chen, Hal Daumé III, Jesse Dodge, Isabella Duan, Ellie Evans, Felix Friedrich, Avijit Ghosh, Usman Gohar, Sara Hooker, Yacine Jernite, Ria Kalluri, Alberto Lusoli, Alina Leidinger, Michelle Lin, Xiuzhu Lin, Sasha Luccioni, Jennifer Mickel, Margaret Mitchell, Jessica Newman , et al. (6 additional authors not shown)

Abstract: Generative AI systems across modalities, ranging from text (including code), image, audio, and video, have broad social impacts, but there is no official standard for means of evaluating those impacts or for which impacts should be evaluated. In this paper, we present a guide that moves toward a standard approach in evaluating a base generative AI system for any modality in two overarching categor… ▽ More Generative AI systems across modalities, ranging from text (including code), image, audio, and video, have broad social impacts, but there is no official standard for means of evaluating those impacts or for which impacts should be evaluated. In this paper, we present a guide that moves toward a standard approach in evaluating a base generative AI system for any modality in two overarching categories: what can be evaluated in a base system independent of context and what can be evaluated in a societal context. Importantly, this refers to base systems that have no predetermined application or deployment context, including a model itself, as well as system components, such as training data. Our framework for a base system defines seven categories of social impact: bias, stereotypes, and representational harms; cultural values and sensitive content; disparate performance; privacy and data protection; financial costs; environmental costs; and data and content moderation labor costs. Suggested methods for evaluation apply to listed generative modalities and analyses of the limitations of existing evaluations serve as a starting point for necessary investment in future evaluations. We offer five overarching categories for what can be evaluated in a broader societal context, each with its own subcategories: trustworthiness and autonomy; inequality, marginalization, and violence; concentration of authority; labor and creativity; and ecosystem and environment. Each subcategory includes recommendations for mitigating harm. △ Less

Submitted 28 June, 2024; v1 submitted 9 June, 2023; originally announced June 2023.

Comments: This version has been removed by arXiv administrators as the submitter did not have the right to agree to the license at the time of submission

Stronger Together: on the Articulation of Ethical Charters, Legal Tools, and Technical Documentation in ML

Authors: Giada Pistilli, Carlos Munoz Ferrandis, Yacine Jernite, Margaret Mitchell

Abstract: The growing need for accountability of the people behind AI systems can be addressed by leveraging processes in three fields of study: ethics, law, and computer science. While these fields are often considered in isolation, they rely on complementary notions in their interpretation and implementation. In this work, we detail this interdependence and motivate the necessary role of collaborative gov… ▽ More The growing need for accountability of the people behind AI systems can be addressed by leveraging processes in three fields of study: ethics, law, and computer science. While these fields are often considered in isolation, they rely on complementary notions in their interpretation and implementation. In this work, we detail this interdependence and motivate the necessary role of collaborative governance tools in shaping a positive evolution of AI. We first contrast notions of compliance in the ethical, legal, and technical fields; we outline both their differences and where they complement each other, with a particular focus on the roles of ethical charters, licenses, and technical documentation in these interactions. We then focus on the role of values in articulating the synergies between the fields and outline specific mechanisms of interaction between them in practice. We identify how these mechanisms have played out in several open governance fora: an open collaborative workshop, a responsible licensing initiative, and a proposed regulatory framework. By leveraging complementary notions of compliance in these three domains, we can create a more comprehensive framework for governing AI systems that jointly takes into account their technical capabilities, their impact on society, and how technical specifications can inform relevant regulations. Our analysis thus underlines the necessity of joint consideration of the ethical, legal, and technical in AI ethics frameworks to be used on a larger scale to govern AI systems and how the thinking in each of these areas can inform the others. △ Less

Submitted 9 May, 2023; originally announced May 2023.

The ConceptARC Benchmark: Evaluating Understanding and Generalization in the ARC Domain

Authors: Arseny Moskvichev, Victor Vikram Odouard, Melanie Mitchell

Abstract: The abilities to form and abstract concepts is key to human intelligence, but such abilities remain lacking in state-of-the-art AI systems. There has been substantial research on conceptual abstraction in AI, particularly using idealized domains such as Raven's Progressive Matrices and Bongard problems, but even when AI systems succeed on such problems, the systems are rarely evaluated in depth to… ▽ More The abilities to form and abstract concepts is key to human intelligence, but such abilities remain lacking in state-of-the-art AI systems. There has been substantial research on conceptual abstraction in AI, particularly using idealized domains such as Raven's Progressive Matrices and Bongard problems, but even when AI systems succeed on such problems, the systems are rarely evaluated in depth to see if they have actually grasped the concepts they are meant to capture. In this paper we describe an in-depth evaluation benchmark for the Abstraction and Reasoning Corpus (ARC), a collection of few-shot abstraction and analogy problems developed by Chollet [2019]. In particular, we describe ConceptARC, a new, publicly available benchmark in the ARC domain that systematically assesses abstraction and generalization abilities on a number of basic spatial and semantic concepts. ConceptARC differs from the original ARC dataset in that it is specifically organized around "concept groups" -- sets of problems that focus on specific concepts and that are vary in complexity and level of abstraction. We report results on testing humans on this benchmark as well as three machine solvers: the top two programs from a 2021 ARC competition and OpenAI's GPT-4. Our results show that humans substantially outperform the machine solvers on this benchmark, showing abilities to abstract and generalize concepts that are not yet captured by AI systems. We believe that this benchmark will spur improvements in the development of AI systems for conceptual abstraction and in the effective evaluation of such systems. △ Less

Submitted 11 May, 2023; originally announced May 2023.

Journal ref: Transactions on Machine Learning Research, 8/2023

The Roles of Symbols in Neural-based AI: They are Not What You Think!

Authors: Daniel L. Silver, Tom M. Mitchell

Abstract: We propose that symbols are first and foremost external communication tools used between intelligent agents that allow knowledge to be transferred in a more efficient and effective manner than having to experience the world directly. But, they are also used internally within an agent through a form of self-communication to help formulate, describe and justify subsymbolic patterns of neural activit… ▽ More We propose that symbols are first and foremost external communication tools used between intelligent agents that allow knowledge to be transferred in a more efficient and effective manner than having to experience the world directly. But, they are also used internally within an agent through a form of self-communication to help formulate, describe and justify subsymbolic patterns of neural activity that truly implement thinking. Symbols, and our languages that make use of them, not only allow us to explain our thinking to others and ourselves, but also provide beneficial constraints (inductive bias) on learning about the world. In this paper we present relevant insights from neuroscience and cognitive science, about how the human brain represents symbols and the concepts they refer to, and how today's artificial neural networks can do the same. We then present a novel neuro-symbolic hypothesis and a plausible architecture for intelligent agents that combines subsymbolic representations for symbols and concepts for learning and reasoning. Our hypothesis and associated architecture imply that symbols will remain critical to the future of intelligent systems NOT because they are the fundamental building blocks of thought, but because they are characterizations of subsymbolic processes that constitute thought. △ Less

Submitted 26 April, 2023; originally announced April 2023.

Comments: 28 pages

Can AI Put Gamma-Ray Astrophysicists Out of a Job?

Authors: Samuel T. Spencer, Vikas Joshi, Alison M. W. Mitchell

Abstract: In what will likely be a litany of generative-model-themed arXiv submissions celebrating April the 1st, we evaluate the capacity of state-of-the-art transformer models to create a paper detailing the detection of a Pulsar Wind Nebula with a non-existent Imaging Atmospheric Cherenkov Telescope (IACT) Array. We do this to evaluate the ability of such models to interpret astronomical observations and… ▽ More In what will likely be a litany of generative-model-themed arXiv submissions celebrating April the 1st, we evaluate the capacity of state-of-the-art transformer models to create a paper detailing the detection of a Pulsar Wind Nebula with a non-existent Imaging Atmospheric Cherenkov Telescope (IACT) Array. We do this to evaluate the ability of such models to interpret astronomical observations and sources based on language information alone, and to assess potential means by which fraudulently generated scientific papers could be identified during peer review (given that reliable generative model watermarking has yet to be deployed for these tools). We conclude that our jobs as astronomers are safe for the time being. From this point on, prompts given to ChatGPT and Stable Diffusion are shown in orange, text generated by ChatGPT is shown in black, whereas analysis by the (human) authors is in blue. △ Less

Submitted 4 April, 2023; v1 submitted 31 March, 2023; originally announced March 2023.

Stable Bias: Analyzing Societal Representations in Diffusion Models

Authors: Alexandra Sasha Luccioni, Christopher Akiki, Margaret Mitchell, Yacine Jernite

Abstract: As machine learning-enabled Text-to-Image (TTI) systems are becoming increasingly prevalent and seeing growing adoption as commercial services, characterizing the social biases they exhibit is a necessary first step to lowering their risk of discriminatory outcomes. This evaluation, however, is made more difficult by the synthetic nature of these systems' outputs: common definitions of diversity a… ▽ More As machine learning-enabled Text-to-Image (TTI) systems are becoming increasingly prevalent and seeing growing adoption as commercial services, characterizing the social biases they exhibit is a necessary first step to lowering their risk of discriminatory outcomes. This evaluation, however, is made more difficult by the synthetic nature of these systems' outputs: common definitions of diversity are grounded in social categories of people living in the world, whereas the artificial depictions of fictive humans created by these systems have no inherent gender or ethnicity. To address this need, we propose a new method for exploring the social biases in TTI systems. Our approach relies on characterizing the variation in generated images triggered by enumerating gender and ethnicity markers in the prompts, and comparing it to the variation engendered by spanning different professions. This allows us to (1) identify specific bias trends, (2) provide targeted scores to directly compare models in terms of diversity and representation, and (3) jointly model interdependent social variables to support a multidimensional analysis. We leverage this method to analyze images generated by 3 popular TTI systems (Dall-E 2, Stable Diffusion v 1.4 and 2) and find that while all of their outputs show correlations with US labor demographics, they also consistently under-represent marginalized identities to different extents. We also release the datasets and low-code interactive bias exploration platforms developed for this work, as well as the necessary tools to similarly evaluate additional TTI systems. △ Less

Submitted 9 November, 2023; v1 submitted 20 March, 2023; originally announced March 2023.

Comments: Accepted to NeurIPS Datasets and Benchmarks 2023 (spotlight)

The BigScience ROOTS Corpus: A 1.6TB Composite Multilingual Dataset

Authors: Hugo Laurençon, Lucile Saulnier, Thomas Wang, Christopher Akiki, Albert Villanova del Moral, Teven Le Scao, Leandro Von Werra, Chenghao Mou, Eduardo González Ponferrada, Huu Nguyen, Jörg Frohberg, Mario Šaško, Quentin Lhoest, Angelina McMillan-Major, Gerard Dupont, Stella Biderman, Anna Rogers, Loubna Ben allal, Francesco De Toni, Giada Pistilli, Olivier Nguyen, Somaieh Nikpoor, Maraim Masoud, Pierre Colombo, Javier de la Rosa , et al. (29 additional authors not shown)

Abstract: As language models grow ever larger, the need for large-scale high-quality text datasets has never been more pressing, especially in multilingual settings. The BigScience workshop, a 1-year international and multidisciplinary initiative, was formed with the goal of researching and training large language models as a values-driven undertaking, putting issues of ethics, harm, and governance in the f… ▽ More As language models grow ever larger, the need for large-scale high-quality text datasets has never been more pressing, especially in multilingual settings. The BigScience workshop, a 1-year international and multidisciplinary initiative, was formed with the goal of researching and training large language models as a values-driven undertaking, putting issues of ethics, harm, and governance in the foreground. This paper documents the data creation and curation efforts undertaken by BigScience to assemble the Responsible Open-science Open-collaboration Text Sources (ROOTS) corpus, a 1.6TB dataset spanning 59 languages that was used to train the 176-billion-parameter BigScience Large Open-science Open-access Multilingual (BLOOM) language model. We further release a large initial subset of the corpus and analyses thereof, and hope to empower large-scale monolingual and multilingual modeling projects with both the data and the processing tools, as well as stimulate research around this large multilingual corpus. △ Less

Submitted 7 March, 2023; originally announced March 2023.

Comments: NeurIPS 2022, Datasets and Benchmarks Track

ACM Class: I.2.7

Read and Reap the Rewards: Learning to Play Atari with the Help of Instruction Manuals

Authors: Yue Wu, Yewen Fan, Paul Pu Liang, Amos Azaria, Yuanzhi Li, Tom M. Mitchell

Abstract: High sample complexity has long been a challenge for RL. On the other hand, humans learn to perform tasks not only from interaction or demonstrations, but also by reading unstructured text documents, e.g., instruction manuals. Instruction manuals and wiki pages are among the most abundant data that could inform agents of valuable features and policies or task-specific environmental dynamics and re… ▽ More High sample complexity has long been a challenge for RL. On the other hand, humans learn to perform tasks not only from interaction or demonstrations, but also by reading unstructured text documents, e.g., instruction manuals. Instruction manuals and wiki pages are among the most abundant data that could inform agents of valuable features and policies or task-specific environmental dynamics and reward structures. Therefore, we hypothesize that the ability to utilize human-written instruction manuals to assist learning policies for specific tasks should lead to a more efficient and better-performing agent. We propose the Read and Reward framework. Read and Reward speeds up RL algorithms on Atari games by reading manuals released by the Atari game developers. Our framework consists of a QA Extraction module that extracts and summarizes relevant information from the manual and a Reasoning module that evaluates object-agent interactions based on information from the manual. An auxiliary reward is then provided to a standard A2C RL agent, when interaction is detected. Experimentally, various RL algorithms obtain significant improvement in performance and training speed when assisted by our design. △ Less

Submitted 20 July, 2024; v1 submitted 9 February, 2023; originally announced February 2023.

Measuring Data

Authors: Margaret Mitchell, Alexandra Sasha Luccioni, Nathan Lambert, Marissa Gerchick, Angelina McMillan-Major, Ezinwanne Ozoani, Nazneen Rajani, Tristan Thrush, Yacine Jernite, Douwe Kiela

Abstract: We identify the task of measuring data to quantitatively characterize the composition of machine learning data and datasets. Similar to an object's height, width, and volume, data measurements quantify different attributes of data along common dimensions that support comparison. Several lines of research have proposed what we refer to as measurements, with differing terminology; we bring some of t… ▽ More We identify the task of measuring data to quantitatively characterize the composition of machine learning data and datasets. Similar to an object's height, width, and volume, data measurements quantify different attributes of data along common dimensions that support comparison. Several lines of research have proposed what we refer to as measurements, with differing terminology; we bring some of this work together, particularly in fields of computer vision and language, and build from it to motivate measuring data as a critical component of responsible AI development. Measuring data aids in systematically building and analyzing machine learning (ML) data towards specific goals and gaining better control of what modern ML systems will learn. We conclude with a discussion of the many avenues of future work, the limitations of data measurements, and how to leverage these measurement approaches in research and practice. △ Less

Submitted 13 February, 2023; v1 submitted 9 December, 2022; originally announced December 2022.

The Stack: 3 TB of permissively licensed source code

Authors: Denis Kocetkov, Raymond Li, Loubna Ben Allal, Jia Li, Chenghao Mou, Carlos Muñoz Ferrandis, Yacine Jernite, Margaret Mitchell, Sean Hughes, Thomas Wolf, Dzmitry Bahdanau, Leandro von Werra, Harm de Vries

Abstract: Large Language Models (LLMs) play an ever-increasing role in the field of Artificial Intelligence (AI)--not only for natural language processing but also for code understanding and generation. To stimulate open and responsible research on LLMs for code, we introduce The Stack, a 3.1 TB dataset consisting of permissively licensed source code in 30 programming languages. We describe how we collect t… ▽ More Large Language Models (LLMs) play an ever-increasing role in the field of Artificial Intelligence (AI)--not only for natural language processing but also for code understanding and generation. To stimulate open and responsible research on LLMs for code, we introduce The Stack, a 3.1 TB dataset consisting of permissively licensed source code in 30 programming languages. We describe how we collect the full dataset, construct a permissively licensed subset, present a data governance plan, discuss limitations, and show promising results on text2code benchmarks by training 350M-parameter decoders on different Python subsets. We find that (1) near-deduplicating the data significantly boosts performance across all experiments, and (2) it is possible to match previously reported HumanEval and MBPP performance using only permissively licensed data. We make the dataset available at https://hf.co/BigCode, provide a tool called "Am I in The Stack" (https://hf.co/spaces/bigcode/in-the-stack) for developers to search The Stack for copies of their code, and provide a process for code to be removed from the dataset by following the instructions at https://www.bigcode-project.org/docs/about/the-stack/. △ Less

Submitted 20 November, 2022; originally announced November 2022.

BLOOM: A 176B-Parameter Open-Access Multilingual Language Model

Authors: BigScience Workshop, :, Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ilić, Daniel Hesslow, Roman Castagné, Alexandra Sasha Luccioni, François Yvon, Matthias Gallé, Jonathan Tow, Alexander M. Rush, Stella Biderman, Albert Webson, Pawan Sasanka Ammanamanchi, Thomas Wang, Benoît Sagot, Niklas Muennighoff, Albert Villanova del Moral, Olatunji Ruwase, Rachel Bawden, Stas Bekman, Angelina McMillan-Major , et al. (369 additional authors not shown)

Abstract: Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access… ▽ More Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). We find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, we publicly release our models and code under the Responsible AI License. △ Less

Submitted 27 June, 2023; v1 submitted 9 November, 2022; originally announced November 2022.

Gathering Strength, Gathering Storms: The One Hundred Year Study on Artificial Intelligence (AI100) 2021 Study Panel Report

Authors: Michael L. Littman, Ifeoma Ajunwa, Guy Berger, Craig Boutilier, Morgan Currie, Finale Doshi-Velez, Gillian Hadfield, Michael C. Horowitz, Charles Isbell, Hiroaki Kitano, Karen Levy, Terah Lyons, Melanie Mitchell, Julie Shah, Steven Sloman, Shannon Vallor, Toby Walsh

Abstract: In September 2021, the "One Hundred Year Study on Artificial Intelligence" project (AI100) issued the second report of its planned long-term periodic assessment of artificial intelligence (AI) and its impact on society. It was written by a panel of 17 study authors, each of whom is deeply rooted in AI research, chaired by Michael Littman of Brown University. The report, entitled "Gathering Strengt… ▽ More In September 2021, the "One Hundred Year Study on Artificial Intelligence" project (AI100) issued the second report of its planned long-term periodic assessment of artificial intelligence (AI) and its impact on society. It was written by a panel of 17 study authors, each of whom is deeply rooted in AI research, chaired by Michael Littman of Brown University. The report, entitled "Gathering Strength, Gathering Storms," answers a set of 14 questions probing critical areas of AI development addressing the major risks and dangers of AI, its effects on society, its public perception and the future of the field. The report concludes that AI has made a major leap from the lab to people's lives in recent years, which increases the urgency to understand its potential negative effects. The questions were developed by the AI100 Standing Committee, chaired by Peter Stone of the University of Texas at Austin, consisting of a group of AI leaders with expertise in computer science, sociology, ethics, economics, and other disciplines. △ Less

Submitted 27 October, 2022; originally announced October 2022.

Comments: 82 pages, https://ai100.stanford.edu/gathering-strength-gathering-storms-one-hundred-year-study-artificial-intelligence-ai100-2021-study

The Debate Over Understanding in AI's Large Language Models

Authors: Melanie Mitchell, David C. Krakauer

Abstract: We survey a current, heated debate in the AI research community on whether large pre-trained language models can be said to "understand" language -- and the physical and social situations language encodes -- in any important sense. We describe arguments that have been made for and against such understanding, and key questions for the broader sciences of intelligence that have arisen in light of th… ▽ More We survey a current, heated debate in the AI research community on whether large pre-trained language models can be said to "understand" language -- and the physical and social situations language encodes -- in any important sense. We describe arguments that have been made for and against such understanding, and key questions for the broader sciences of intelligence that have arisen in light of these arguments. We contend that a new science of intelligence can be developed that will provide insight into distinct modes of understanding, their strengths and limitations, and the challenge of integrating diverse forms of cognition. △ Less

Submitted 10 February, 2023; v1 submitted 14 October, 2022; originally announced October 2022.

Comments: Under submission as a Perspective article. Updated with additional discussion and citations

Journal ref: Proceedings of the National Academy of Sciences 120 (13), 2023

Embodied, Situated, and Grounded Intelligence: Implications for AI

Authors: Tyler Millhouse, Melanie Moses, Melanie Mitchell

Abstract: In April of 2022, the Santa Fe Institute hosted a workshop on embodied, situated, and grounded intelligence as part of the Institute's Foundations of Intelligence project. The workshop brought together computer scientists, psychologists, philosophers, social scientists, and others to discuss the science of embodiment and related issues in human intelligence, and its implications for building robus… ▽ More In April of 2022, the Santa Fe Institute hosted a workshop on embodied, situated, and grounded intelligence as part of the Institute's Foundations of Intelligence project. The workshop brought together computer scientists, psychologists, philosophers, social scientists, and others to discuss the science of embodiment and related issues in human intelligence, and its implications for building robust, human-level AI. In this report, we summarize each of the talks and the subsequent discussions. We also draw out a number of key themes and identify important frontiers for future research. △ Less

Submitted 24 October, 2022; originally announced October 2022.

Comments: 38 pages, workshop report

SEAL : Interactive Tool for Systematic Error Analysis and Labeling

Authors: Nazneen Rajani, Weixin Liang, Lingjiao Chen, Meg Mitchell, James Zou

Abstract: With the advent of Transformers, large language models (LLMs) have saturated well-known NLP benchmarks and leaderboards with high aggregate performance. However, many times these models systematically fail on tail data or rare groups not obvious in aggregate evaluation. Identifying such problematic data groups is even more challenging when there are no explicit labels (e.g., ethnicity, gender, etc… ▽ More With the advent of Transformers, large language models (LLMs) have saturated well-known NLP benchmarks and leaderboards with high aggregate performance. However, many times these models systematically fail on tail data or rare groups not obvious in aggregate evaluation. Identifying such problematic data groups is even more challenging when there are no explicit labels (e.g., ethnicity, gender, etc.) and further compounded for NLP datasets due to the lack of visual features to characterize failure modes (e.g., Asian males, animals indoors, waterbirds on land, etc.). This paper introduces an interactive Systematic Error Analysis and Labeling (\seal) tool that uses a two-step approach to first identify high error slices of data and then, in the second step, introduce methods to give human-understandable semantics to those underperforming slices. We explore a variety of methods for coming up with coherent semantics for the error groups using language models for semantic labeling and a text-to-image model for generating visual features. SEAL toolkit and demo screencast is available at https://huggingface.co/spaces/nazneen/seal. △ Less

Submitted 11 October, 2022; originally announced October 2022.

Comments: Accepted at EMNLP 2022 demo track

A Human Rights-Based Approach to Responsible AI

Authors: Vinodkumar Prabhakaran, Margaret Mitchell, Timnit Gebru, Iason Gabriel

Abstract: Research on fairness, accountability, transparency and ethics of AI-based interventions in society has gained much-needed momentum in recent years. However it lacks an explicit alignment with a set of normative values and principles that guide this research and interventions. Rather, an implicit consensus is often assumed to hold for the values we impart into our models - something that is at odds… ▽ More Research on fairness, accountability, transparency and ethics of AI-based interventions in society has gained much-needed momentum in recent years. However it lacks an explicit alignment with a set of normative values and principles that guide this research and interventions. Rather, an implicit consensus is often assumed to hold for the values we impart into our models - something that is at odds with the pluralistic world we live in. In this paper, we put forth the doctrine of universal human rights as a set of globally salient and cross-culturally recognized set of values that can serve as a grounding framework for explicit value alignment in responsible AI - and discuss its efficacy as a framework for civil society partnership and participation. We argue that a human rights framework orients the research in this space away from the machines and the risks of their biases, and towards humans and the risks to their rights, essentially helping to center the conversation around who is harmed, what harms they face, and how those harms may be mitigated. △ Less

Submitted 6 October, 2022; originally announced October 2022.

Comments: Presented as a (non-archival) poster at the 2022 ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization or (EAAMO '22)

Evaluate & Evaluation on the Hub: Better Best Practices for Data and Model Measurements

Authors: Leandro von Werra, Lewis Tunstall, Abhishek Thakur, Alexandra Sasha Luccioni, Tristan Thrush, Aleksandra Piktus, Felix Marty, Nazneen Rajani, Victor Mustar, Helen Ngo, Omar Sanseviero, Mario Šaško, Albert Villanova, Quentin Lhoest, Julien Chaumond, Margaret Mitchell, Alexander M. Rush, Thomas Wolf, Douwe Kiela

Abstract: Evaluation is a key part of machine learning (ML), yet there is a lack of support and tooling to enable its informed and systematic practice. We introduce Evaluate and Evaluation on the Hub --a set of tools to facilitate the evaluation of models and datasets in ML. Evaluate is a library to support best practices for measurements, metrics, and comparisons of data and models. Its goal is to support… ▽ More Evaluation is a key part of machine learning (ML), yet there is a lack of support and tooling to enable its informed and systematic practice. We introduce Evaluate and Evaluation on the Hub --a set of tools to facilitate the evaluation of models and datasets in ML. Evaluate is a library to support best practices for measurements, metrics, and comparisons of data and models. Its goal is to support reproducibility of evaluation, centralize and document the evaluation process, and broaden evaluation to cover more facets of model performance. It includes over 50 efficient canonical implementations for a variety of domains and scenarios, interactive documentation, and the ability to easily share implementations and outcomes. The library is available at https://github.com/huggingface/evaluate. In addition, we introduce Evaluation on the Hub, a platform that enables the large-scale evaluation of over 75,000 models and 11,000 datasets on the Hugging Face Hub, for free, at the click of a button. Evaluation on the Hub is available at https://huggingface.co/autoevaluate. △ Less

Submitted 6 October, 2022; v1 submitted 30 September, 2022; originally announced October 2022.

Learning Sparsity-Promoting Regularizers using Bilevel Optimization

Authors: Avrajit Ghosh, Michael T. McCann, Madeline Mitchell, Saiprasad Ravishankar

Abstract: We present a method for supervised learning of sparsity-promoting regularizers for denoising signals and images. Sparsity-promoting regularization is a key ingredient in solving modern signal reconstruction problems; however, the operators underlying these regularizers are usually either designed by hand or learned from data in an unsupervised way. The recent success of supervised learning (mainly… ▽ More We present a method for supervised learning of sparsity-promoting regularizers for denoising signals and images. Sparsity-promoting regularization is a key ingredient in solving modern signal reconstruction problems; however, the operators underlying these regularizers are usually either designed by hand or learned from data in an unsupervised way. The recent success of supervised learning (mainly convolutional neural networks) in solving image reconstruction problems suggests that it could be a fruitful approach to designing regularizers. Towards this end, we propose to denoise signals using a variational formulation with a parametric, sparsity-promoting regularizer, where the parameters of the regularizer are learned to minimize the mean squared error of reconstructions on a training set of ground truth image and measurement pairs. Training involves solving a challenging bilievel optimization problem; we derive an expression for the gradient of the training loss using the closed-form solution of the denoising problem and provide an accompanying gradient descent algorithm to minimize it. Our experiments with structured 1D signals and natural images show that the proposed method can learn an operator that outperforms well-known regularizers (total variation, DCT-sparsity, and unsupervised dictionary learning) and collaborative filtering for denoising. While the approach we present is specific to denoising, we believe that it could be adapted to the larger class of inverse problems with linear measurement models, giving it applicability in a wide range of signal reconstruction settings. △ Less

Submitted 5 September, 2023; v1 submitted 18 July, 2022; originally announced July 2022.

Journal ref: SIAM Journal on Imaging Sciences (SIIMS-2023)