Showing 1–39 of 39 results for author: Jose, C

Radio Continuum Emission from Evolving Star-Forming Galaxies -- I. Correlations Involving the Total Synchrotron Luminosity

Authors: Sukanta Ghosh, Luke Chamandy, Charles Jose, Anvar Shukurov, Luiz Felippe S. Rodrigues, Fatemeh Tabatabaei

Abstract: Synchrotron radiation dominates the continuum emission of star-forming galaxies in the frequency range from a few $\rm MHz$ to about $30\,\rm{GHz}$. We model the total synchrotron emission of a large population of evolving star-forming galaxies using the semi-analytic galaxy formation model GALFORM combined with the dynamo simulation code MAGNETIZER. Assuming local energy equipartition between cos… ▽ More Synchrotron radiation dominates the continuum emission of star-forming galaxies in the frequency range from a few $\rm MHz$ to about $30\,\rm{GHz}$. We model the total synchrotron emission of a large population of evolving star-forming galaxies using the semi-analytic galaxy formation model GALFORM combined with the dynamo simulation code MAGNETIZER. Assuming local energy equipartition between cosmic rays and magnetic fields, we calculate the specific synchrotron luminosity $L_ν$ for each simulated galaxy at various frequencies and find strong positive correlations between $L_ν$ and both the star formation rate ($\rm SFR$) and characteristic galaxy rotation speed $V_{\rm rot}$ for redshifts up to $z\simeq 3$. At low redshifts, the turbulent magnetic field is found to dominate in the synchrotron luminosity, but the contribution of the large-scale magnetic field increases with redshift and becomes important for $z\gtrsim 1$. The correlation between $L_ν$ and $\rm SFR$ arises from the tight correlation between the disc gas mass $M_{\rm gas}$ and $\rm SFR$, and the correlation between $L_ν$ and $V_{\rm rot}$ is additionally a consequence of the stellar mass Tully--Fisher relation for main-sequence galaxies. At low redshifts, the model predictions and observational data compiled for this work show remarkable agreement, but a discrepancy arises at higher redshifts, where modelled $\rm SFR$ values are systematically smaller than those previously inferred from observations. These theoretical models will aid the interpretation of next-generation radio surveys with the Square Kilometre Array and other telescopes. △ Less

Submitted 1 June, 2026; originally announced June 2026.

Comments: Main text 35 pages, 30 figures, and 8 tables. Submitted to ApJ. Comments welcome!

AdaDINO: Context-Adaptive DINO-Distilled Vision Foundation Models for Efficient Open-Vocabulary Edge Inference

Authors: Yiwei Zhao, Yi Zheng, Huapeng Su, Jieyu Lin, Stefano Ambrogio, Cijo Jose, Michael Ramamonjisoa, Patrick Labatut, Barbara De Salvo, Chiao Liu, Phillip B. Gibbons, Ziyun Li

Abstract: Always-on contextual AI runs language-aligned vision foundation models (VFMs) on edge devices, where the on-device model is the dominant continuous compute cost under strict latency and power limits. Due to an observed low-frequency shift in scene context and its relevant vocabulary, we present AdaDINO, an adaptive framework that makes on-device VFM inference efficient by matching execution to the… ▽ More Always-on contextual AI runs language-aligned vision foundation models (VFMs) on edge devices, where the on-device model is the dominant continuous compute cost under strict latency and power limits. Due to an observed low-frequency shift in scene context and its relevant vocabulary, we present AdaDINO, an adaptive framework that makes on-device VFM inference efficient by matching execution to the current scene and task. We build on a known phenomenon, that the accuracy drop of shrinking model sizes depends on the task, and turn it into task-level adaptive execution. AdaDINO integrates neural architecture search (NAS) into a language-aligned VFM backbone distilled from DINOv2, training a single family of subnets for efficient execution during runtime. A multimodal large language model (LLM) on the cloud, invoked at low frequency, refines the candidate class set from scene context, while a learned selector activates the least-cost subnet predicted to retain a target fraction of accuracy. With the backbone and semantic pipeline held fixed, learned selection alone reduces average compute by $37\%$ over the best fixed subnet at equal segmentation accuracy. Across zero-shot classification and open-vocabulary segmentation, AdaDINO establishes a strong accuracy-efficiency frontier, improving over evaluated models of comparable sizes by up to $7.9\%$ in acc@1 on IN1K and $5.2\%$ mIoU on ADE20K, and reducing average FLOPs by up to $74.9\%$ at similar accuracy. △ Less

Submitted 1 August, 2026; v1 submitted 16 April, 2026; originally announced April 2026.

The Distribution of Cosmic Ray Electrons in Star-Forming Galaxies

Authors: Anvar Shukurov, Charles Jose

Abstract: We derive explicit, algebraic expressions for the steady-state number density of cosmic ray electrons as a function of position and energy using Green's function of the diffusion equation with energy losses for an axisymmetric distributions of the particle sources in the galactocentric radius $r$ and distance to the mid-plane $z$. The solution is obtained for a Gaussian distribution of the particl… ▽ More We derive explicit, algebraic expressions for the steady-state number density of cosmic ray electrons as a function of position and energy using Green's function of the diffusion equation with energy losses for an axisymmetric distributions of the particle sources in the galactocentric radius $r$ and distance to the mid-plane $z$. The solution is obtained for a Gaussian distribution of the particle sources in $r$ and $z$ but we show that it can be used for an arbitrary spatial distribution of the sources. The accuracy of our results is about 10\% or better in wide ranges of $r$, $z$ and particle energies. These solutions can be used in the interpretation of radio astronomical observations of galaxies, particularly in the studies of the radio luminosities for large galaxy samples, and represent a physically justifiable and efficient alternative to the assumption of the energy equipartition between cosmic rays and interstellar magnetic fields. △ Less

Submitted 26 April, 2026; v1 submitted 5 April, 2026; originally announced April 2026.

Comments: 20 pages, 4 figures, submitted as research topic `The Role of Plasmas and Cosmic Magnetism in High-energy Astroparticle Physics' to Frontiers in Astronomy and Space Sciences. Revised after referees' comments

Efficient Universal Perception Encoder

Authors: Chenchen Zhu, Saksham Suri, Cijo Jose, Maxime Oquab, Marc Szafraniec, Wei Wen, Yunyang Xiong, Patrick Labatut, Piotr Bojanowski, Raghuraman Krishnamoorthi, Vikas Chandra

Abstract: Running AI models on smart edge devices can unlock versatile user experiences, but presents challenges due to limited compute and the need to handle multiple tasks simultaneously. This requires a vision encoder with small size but powerful and versatile representations. We present our method, Efficient Universal Perception Encoder (EUPE), which offers both inference efficiency and universally good… ▽ More Running AI models on smart edge devices can unlock versatile user experiences, but presents challenges due to limited compute and the need to handle multiple tasks simultaneously. This requires a vision encoder with small size but powerful and versatile representations. We present our method, Efficient Universal Perception Encoder (EUPE), which offers both inference efficiency and universally good representations for diverse downstream tasks. We achieve this by distilling from multiple domain-expert foundation vision encoders. Unlike previous agglomerative methods that directly scale down from multiple teachers to an efficient encoder, we demonstrate the importance of first scaling up to a large proxy teacher and then scaling down from this single teacher. Experiments show that EUPE achieves on-par or better performance than individual domain experts of the same size on diverse task domains and also outperforms previous agglomerative encoders. We release the full family of EUPE models and the code to foster future research. △ Less

Submitted 31 March, 2026; v1 submitted 23 March, 2026; originally announced March 2026.

Comments: Code: https://github.com/facebookresearch/EUPE; Model: https://huggingface.co/collections/facebook/eupe

Perspiration vapor lightens near skin air but hinders human evaporative cooling in arid heat

Authors: Shri H. Viswanathan, Ankit Joshi, Isabella DeClair, Bryce Twidwell, Muhammad Abdullah, Lyle Bartels, Faisal Abedin, Joseph Rotella, Cibin T. Jose, Konrad Rykaczewski

Abstract: Sweat evaporation is the body's primary cooling mechanism, yet the physical factors governing it are not fully understood. We identify a dueling buoyancy effect in the context of the human body, in which perspiration vapor reduces the near skin air density, counteracting the downward flow driven by cooling of warm air upon contact with the skin. In hot, arid, stagnant environments, this opposing b… ▽ More Sweat evaporation is the body's primary cooling mechanism, yet the physical factors governing it are not fully understood. We identify a dueling buoyancy effect in the context of the human body, in which perspiration vapor reduces the near skin air density, counteracting the downward flow driven by cooling of warm air upon contact with the skin. In hot, arid, stagnant environments, this opposing buoyancy suppresses free convection and can reduce sweat evaporation by more than half. As a result, commonly used thermoregulation models can substantially underpredict body temperature (e.g., by 1C after 2 hours of exposure to typical Arizona summer conditions). We develop compact, physics informed models for free convective heat transfer coefficients across wide temperature and humidity ranges, enabling improved thermoregulation modeling and thermal audits. These results enhance understanding of human heat balance and support more accurate heat stress assessment to inform behavioral, infrastructural, and policy decisions for extreme heat adaptations. △ Less

Submitted 7 May, 2026; v1 submitted 20 November, 2025; originally announced November 2025.

Comments: 18 pages, 4 figures and supporting information

DINOv3

Authors: Oriane Siméoni, Huy V. Vo, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Michaël Ramamonjisoa, Francisco Massa, Daniel Haziza, Luca Wehrstedt, Jianyuan Wang, Timothée Darcet, Théo Moutakanni, Leonel Sentana, Claire Roberts, Andrea Vedaldi, Jamie Tolan, John Brandt, Camille Couprie, Julien Mairal, Hervé Jégou, Patrick Labatut , et al. (1 additional authors not shown)

Abstract: Self-supervised learning holds the promise of eliminating the need for manual data annotation, enabling models to scale effortlessly to massive datasets and larger architectures. By not being tailored to specific tasks or domains, this training paradigm has the potential to learn visual representations from diverse sources, ranging from natural to aerial images -- using a single algorithm. This te… ▽ More Self-supervised learning holds the promise of eliminating the need for manual data annotation, enabling models to scale effortlessly to massive datasets and larger architectures. By not being tailored to specific tasks or domains, this training paradigm has the potential to learn visual representations from diverse sources, ranging from natural to aerial images -- using a single algorithm. This technical report introduces DINOv3, a major milestone toward realizing this vision by leveraging simple yet effective strategies. First, we leverage the benefit of scaling both dataset and model size by careful data preparation, design, and optimization. Second, we introduce a new method called Gram anchoring, which effectively addresses the known yet unsolved issue of dense feature maps degrading during long training schedules. Finally, we apply post-hoc strategies that further enhance our models' flexibility with respect to resolution, model size, and alignment with text. As a result, we present a versatile vision foundation model that outperforms the specialized state of the art across a broad range of settings, without fine-tuning. DINOv3 produces high-quality dense features that achieve outstanding performance on various vision tasks, significantly surpassing previous self- and weakly-supervised foundation models. We also share the DINOv3 suite of vision models, designed to advance the state of the art on a wide spectrum of tasks and data by providing scalable solutions for diverse resource constraints and deployment scenarios. △ Less

Submitted 13 August, 2025; originally announced August 2025.

AGTCNet: A Graph-Temporal Approach for Principled Motor Imagery EEG Classification

Authors: Galvin Brice S. Lim, Brian Godwin S. Lim, Argel A. Bandala, John Anthony C. Jose, Timothy Scott C. Chu, Edwin Sybingco

Abstract: Brain-computer interface (BCI) technology utilizing electroencephalography (EEG) marks a transformative innovation, empowering motor-impaired individuals to engage with their environment on equal footing. Despite its promising potential, developing subject-invariant and session-invariant BCI systems remains a significant challenge due to the inherent complexity and variability of neural activity a… ▽ More Brain-computer interface (BCI) technology utilizing electroencephalography (EEG) marks a transformative innovation, empowering motor-impaired individuals to engage with their environment on equal footing. Despite its promising potential, developing subject-invariant and session-invariant BCI systems remains a significant challenge due to the inherent complexity and variability of neural activity across individuals and over time, compounded by EEG hardware constraints. While prior studies have sought to develop robust BCI systems, existing approaches remain ineffective in capturing the intricate spatiotemporal dependencies within multichannel EEG signals. This study addresses this gap by introducing the attentive graph-temporal convolutional network (AGTCNet), a novel graph-temporal model for motor imagery EEG (MI-EEG) classification. Specifically, AGTCNet leverages the topographic configuration of EEG electrodes as an inductive bias and integrates graph convolutional attention network (GCAT) to jointly learn expressive spatiotemporal EEG representations. The proposed model significantly outperformed existing MI-EEG classifiers, achieving state-of-the-art performance while utilizing a compact architecture, underscoring its effectiveness and practicality for BCI deployment. With a 49.87% reduction in model size, 64.65% faster inference time, and shorter input EEG signal, AGTCNet achieved a moving average accuracy of 66.82% for subject-independent classification on the BCI Competition IV Dataset 2a, which further improved to 82.88% when fine-tuned for subject-specific classification. On the EEG Motor Movement/Imagery Dataset, AGTCNet achieved moving average accuracies of 64.14% and 85.22% for 4-class and 2-class subject-independent classifications, respectively, with further improvements to 72.13% and 90.54% for subject-specific classifications. △ Less

Submitted 26 June, 2025; originally announced June 2025.

Comments: This work has been submitted to the IEEE for possible publication

Journal ref: IEEE Access. 13(2025) 187383-187409

Non-Separable Halo Bias from High-Redshift Galaxy Clustering

Authors: Emy Mons, Vipul Prasad Maranchery, M. S. Suryan Sivadas, Charles Jose

Abstract: The halo model provides a powerful framework for interpreting galaxy clustering by linking the spatial distribution of dark matter haloes to the underlying matter distribution. A key assumption within the halo bias approximation of the halo model is that, on sufficiently large scales, the halo bias between two halo populations is a separable function of the mass of each population. In this work, w… ▽ More The halo model provides a powerful framework for interpreting galaxy clustering by linking the spatial distribution of dark matter haloes to the underlying matter distribution. A key assumption within the halo bias approximation of the halo model is that, on sufficiently large scales, the halo bias between two halo populations is a separable function of the mass of each population. In this work, we test the validity of this approximation on quasi-linear scales using both simulations and observational data across a broad range of halo masses and redshifts. In particular, we define a separability function based on halo or galaxy cross-correlations to quantify deviations from halo bias separability, and measure it from N-body simulations. We find significant departures from separability on quasi-linear scales (\(\sim 1\text{--}5\,\mathrm{Mpc}\)) at high redshifts (\(z \geq 3\)), leading to a suppression in the scale-dependent halo bias and hence in halo cross-correlations by up to a factor of 2 -- or even higher. In contrast, deviations at low redshifts remain modest. Additionally, using high-redshift (\(z \sim 3.6\)) galaxy samples, we detect deviations from bias separability that closely align with simulation predictions. The breakdown of the separable bias approximation on quasi-linear scales at high redshifts underscore the importance to account for non-separability in models of the galaxy-halo connection in this regime. Furthermore, these results highlight the potential of high-redshift galaxy cross-correlations as a probe for improving the galaxy-halo connection from upcoming large-scale surveys. △ Less

Submitted 30 January, 2026; v1 submitted 9 June, 2025; originally announced June 2025.

Comments: 11 pages, 5 figures

A micro-to-macroscale and multi-method investigation of human sweating dynamics

Authors: Cibin T. Jose, Ankit Joshi, Shri H. Viswanathan, Sincere K. Nash, Kambiz Sadeghi, Stavros A. Kavouras, Konrad Rykaczewski

Abstract: Sweat secretion and evaporation from the skin dictate the human ability to thermoregulate and thermal comfort in hot environments and impact skin interactions with cosmetics, textiles, and wearable electronics or sensors. However, sweating has mostly been investigated using macroscopic physiological methods, leaving micro-to-macroscale sweating dynamics unexplored. We explore these processes by em… ▽ More Sweat secretion and evaporation from the skin dictate the human ability to thermoregulate and thermal comfort in hot environments and impact skin interactions with cosmetics, textiles, and wearable electronics or sensors. However, sweating has mostly been investigated using macroscopic physiological methods, leaving micro-to-macroscale sweating dynamics unexplored. We explore these processes by employing a coupled microscale imaging and transport measurement approach used in engineering studies of phase change processes. Specifically, we employed a comprehensive set of macroscale physiological measurements (ventilated capsule sweat rate, galvanic skin conductance, and dielectric epidermis hydration) complemented by three microscale imaging techniques (visible light, midwave infrared, and optical coherence tomography imaging). Inspired by industrial jet cooling devices, we also explore an air jet (vs. cylindrical) capsule for measuring sweat rate. To enable near simultaneous application of these methods, we studied forehead sweating dynamics of six supine subjects undergoing passive heating, cooling, and secondary heating. The relative dynamics of the physiological measurements agree with prior observations and can be explained using imaged microscale sweating dynamics. This comprehensive study provides new insights into the biophysical dynamics of sweating onset and following cyclic porewise, transition, and filmwise sweating modes, and highlights the roles of stratum corneum hydration, salt deposits, and microscale hair. △ Less

Submitted 7 May, 2025; originally announced May 2025.

Comments: 30 pages, 8 figures

Resolving shortwave and longwave irradiation distributions across the human body in outdoor built environments

Authors: Kambiz Sadeghi, Shri H. Viswanathan, Ankit Joshi, Lyle Bartels, Sylwester Wereski, Cibin T. Jose, Galina Mihaleva, Muhammad Abdullah, Ariane Middel, Konrad Rykaczewski

Abstract: Outdoor built environments can be designed to enhance thermal comfort, yet the relationship between the two is often assessed in whole-body terms, overlooking the asymmetric nature of thermal interactions between the human body and its surroundings. Moreover, the radiative component of heat exchange-dominant in hot and dry climates-is typically lumped into a single artificial metric, the mean radi… ▽ More Outdoor built environments can be designed to enhance thermal comfort, yet the relationship between the two is often assessed in whole-body terms, overlooking the asymmetric nature of thermal interactions between the human body and its surroundings. Moreover, the radiative component of heat exchange-dominant in hot and dry climates-is typically lumped into a single artificial metric, the mean radiant temperature, rather than being resolved into its shortwave and longwave spectral components. The shortwave irradiation distribution on the human body is often highly anisotropic, causing localized thermal discomfort in outdoor environments. However, no existing methods effectively quantify shortwave and longwave irradiation distributions on the human body. To address this gap, we developed two methods to quantify these processes. The first approach uses an outdoor thermal manikin with a white-coated side, enabling the separation of spectral components by subtracting measurements from symmetrically corresponding surface zones of tan color. The second hybrid approach converts radiometer measurements in six directions into boundary conditions for computational thermal manikin simulations. We evaluated irradiation distributions for various body parts using both methods during outdoor measurements across sunny, partially shaded, and fully shaded sites under warm to extremely hot conditions. In most cases, the two methods produced closely aligned results, with divergences highlighting their respective strengths and limitations. Additionally, we used the manikin to quantify irradiation attenuation provided by five long-sleeve shirts with colors ranging from white to black. These advanced methods can be integrated with airflow and thermoregulatory modeling to optimize outdoor built environments for enhanced human thermal comfort. △ Less

Submitted 12 March, 2025; v1 submitted 6 February, 2025; originally announced February 2025.

Comments: 29 pages, 10 figures

Journal ref: Building and Environment, 2025, 112934, ISSN 0360-1323

DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment

Authors: Cijo Jose, Théo Moutakanni, Dahyun Kang, Federico Baldassarre, Timothée Darcet, Hu Xu, Daniel Li, Marc Szafraniec, Michaël Ramamonjisoa, Maxime Oquab, Oriane Siméoni, Huy V. Vo, Patrick Labatut, Piotr Bojanowski

Abstract: Self-supervised visual foundation models produce powerful embeddings that achieve remarkable performance on a wide range of downstream tasks. However, unlike vision-language models such as CLIP, self-supervised visual features are not readily aligned with language, hindering their adoption in open-vocabulary tasks. Our method, named dino.txt, unlocks this new ability for DINOv2, a widely used self… ▽ More Self-supervised visual foundation models produce powerful embeddings that achieve remarkable performance on a wide range of downstream tasks. However, unlike vision-language models such as CLIP, self-supervised visual features are not readily aligned with language, hindering their adoption in open-vocabulary tasks. Our method, named dino.txt, unlocks this new ability for DINOv2, a widely used self-supervised visual encoder. We build upon the LiT training strategy, which trains a text encoder to align with a frozen vision model but leads to unsatisfactory results on dense tasks. We propose several key ingredients to improve performance on both global and dense tasks, such as concatenating the [CLS] token with the patch average to train the alignment and curating data using both text and image modalities. With these, we successfully train a CLIP-like model with only a fraction of the computational cost compared to CLIP while achieving state-of-the-art results in zero-shot classification and open-vocabulary semantic segmentation. △ Less

Submitted 20 December, 2024; originally announced December 2024.

Probing Environmental Dependence of High-Redshift Galaxy Properties with the Marked Correlation Function

Authors: Emy Mons, Charles Jose

Abstract: In hierarchical structure formation, correlations between galaxy properties and their environments reveal important clues about galaxy evolution, emphasizing the importance of measuring these relationships. We probe the environmental dependence of Lyman-break galaxy (LBG) properties in the redshift range of $3$ to $5$ using marked correlation function statistics with galaxy samples from the Hyper… ▽ More In hierarchical structure formation, correlations between galaxy properties and their environments reveal important clues about galaxy evolution, emphasizing the importance of measuring these relationships. We probe the environmental dependence of Lyman-break galaxy (LBG) properties in the redshift range of $3$ to $5$ using marked correlation function statistics with galaxy samples from the Hyper Suprime-Cam Subaru Strategic Program and the Canada--France--Hawaii Telescope U-band surveys. We find that the UV magnitude and color of magnitude-selected LBG samples are strongly correlated with their environment, making these properties effective tracers of it. In contrast, the star formation rate and stellar mass of LBGs exhibit a weak environmental dependence. For UV magnitudes and color, the correlation is stronger in brighter galaxy samples across all redshifts and extends to scales far beyond the size of typical dark matter halos. This suggests that within a given sample, LBGs with high UV magnitudes or colors are more likely to form pairs at these scales than predicted by the two-point angular correlation function. Moreover, the amplitude of the marked correlation function is generally higher for LBG samples compared to that of $z \sim 0$ galaxies from previous studies.We also find that for LBG samples selected by the same absolute threshold magnitude or average halo mass, the correlation between UV magnitudes and the environment generally becomes more pronounced as the redshift decreases. On the other hand, for samples with the same effective large-scale bias at $z\sim 4$ and $5$, the marked correlation functions are similar on large scales. △ Less

Submitted 8 January, 2025; v1 submitted 17 December, 2024; originally announced December 2024.

Comments: 13 pages, version submitted to The Open Journal of Astrophysics

Journal ref: The Open Journal of Astrophysics , Vol 8 (January 2025)

Pressure-Induced Phase Transitions in Bilayer La$_3$Ni$_2$O$_7$

Authors: Mingyu Xu, Greeshma C. Jose, Aya Rutherford, Haozhe Wang, Stephen Zhang, Robert J. Cava, Haidong Zhou, Wenli Bi, Weiwei Xie

Abstract: La$_3$Ni$_2$O$_7$ exists in two polymorphs: an unconventional structure with alternating layers of single- and triple-layered nickel-oxygen octahedra, and a classical double-layered Ruddlesden-Popper phase. In this study, we report the growth of single crystals of classical double-layered La$_3$Ni$_2$O$_7$ using the floating zone method. Structural characterization under pressures up to 15.4 GPa r… ▽ More La$_3$Ni$_2$O$_7$ exists in two polymorphs: an unconventional structure with alternating layers of single- and triple-layered nickel-oxygen octahedra, and a classical double-layered Ruddlesden-Popper phase. In this study, we report the growth of single crystals of classical double-layered La$_3$Ni$_2$O$_7$ using the floating zone method. Structural characterization under pressures up to 15.4 GPa reveals a gradual transition from orthorhombic to tetragonal symmetry near 12 GPa. Additionally, we present pressure and field-dependent electrical resistance measurements under pressures as high as 27.4 GPa, from which we construct a phase diagram. △ Less

Submitted 24 October, 2024; originally announced October 2024.

Comments: 13+5 pages, 4+2 figures

Pressure Tuning the Mixture of Eu$^{2+}$ and Eu$^{3+}$ in Eu$_4$Bi$_6$Se$_{13}$

Authors: Mingyu Xu, Jose L. Gonzalez Jimenez, Greeshma C. Jose, Artittaya Boonkird, Chengkun Xing, Chelsea Harrod, Xinle Li, Haidong Zhou, Alyssa Gaiser, Xianglin Ke, Wenli Bi, Mingda Li, Weiwei Xie

Abstract: The investigation of crystallographic, electronic, and magnetic characteristics, especially the mixed valences of Eu$^{2+}$ and Eu$^{3+}$ under pressure of a novel europium-based bismuth selenide compound, Eu$_4$Bi$_6$Se$_{13}$, presented. This new compound adopts a monoclinic crystal structure classified under the P$2_1$/m space group (#11). It exhibits distinctive structural features, including… ▽ More The investigation of crystallographic, electronic, and magnetic characteristics, especially the mixed valences of Eu$^{2+}$ and Eu$^{3+}$ under pressure of a novel europium-based bismuth selenide compound, Eu$_4$Bi$_6$Se$_{13}$, presented. This new compound adopts a monoclinic crystal structure classified under the P$2_1$/m space group (#11). It exhibits distinctive structural features, including substantial Eu-Se coordination numbers, Bi-Se ladders, and linear chains of Eu atoms that propagate along the b-axis. Electronic resistivity assessments indicate that Eu$_{4}$Bi$_{6}$Se$_{13}$ exhibits weak metallic behaviors. Magnetic characterization reveals uniaxial magnetic anisotropy, with a notable spin transition at approximately 1.2 T when the magnetic field is oriented along the b-axis. This behavior, coupled with the specific Eu-Eu interatomic distances and the magnetic saturation observed at low fields, supports the identification of metamagnetic properties attributable to the flipping of europium spins. The Curie-Weiss analysis of the magnetic susceptibility measured both perpendicular and parallel to the b-axis and high-pressure partial fluorescence yield (PFY) results detected by X-ray absorption spectroscopy (XAS) reveal the tendency of the material to enter a mixed valent state where the trivalent state becomes more prominent with the pressure increase or temperature decrease. △ Less

Submitted 28 June, 2024; originally announced July 2024.

Comments: 22 pages 8 figures

Understanding the radio luminosity function of star-forming galaxies and its cosmological evolution

Authors: Charles Jose, Luke Chamandy, Anvar Shukurov, Kandaswamy Subramanian, Luiz Felippe S. Rodrigues, Carlton M. Baugh

Abstract: We explore the redshift evolution of the radio luminosity function (RLF) of star-forming galaxies using GALFORM, a semi-analytic model of galaxy formation and a dynamo model of the magnetic field evolving in a galaxy. Assuming energy equipartition between the magnetic field and cosmic rays, we derive the synchrotron luminosity of each sample galaxy. In a model where the turbulent speed is correlat… ▽ More We explore the redshift evolution of the radio luminosity function (RLF) of star-forming galaxies using GALFORM, a semi-analytic model of galaxy formation and a dynamo model of the magnetic field evolving in a galaxy. Assuming energy equipartition between the magnetic field and cosmic rays, we derive the synchrotron luminosity of each sample galaxy. In a model where the turbulent speed is correlated with the star formation rate, the RLF is in fair agreement with observations in the redshift range $0 \leq z \leq 2$. At larger redshifts, the structure of galaxies, their interstellar matter and turbulence appear to be rather different from those at $z\lesssim2$, so that the turbulence and magnetic field models applicable at low redshifts become inadequate. The strong redshift evolution of the RLF at $0 \leq z \leq 2$ can be attributed to an increased number, at high redshift, of galaxies with large disc volumes and strong magnetic fields. On the other hand, in models where the turbulent speed is a constant or an explicit function of $z$, the observed redshift evolution of the RLF is poorly captured. The evolution of the interstellar turbulence and outflow parameters appear to be major (but not the only) drivers of the RLF changes. We find that both the small- and large-scale magnetic fields contribute to the RLF but the small-scale field dominates at high redshifts. Polarisation observations will therefore be important to distinguish these two components and understand better the evolution of galaxies and their nonthermal constituents. △ Less

Submitted 11 June, 2024; v1 submitted 23 February, 2024; originally announced February 2024.

Comments: 19 pages, 8 figures. MNRAS, in press

Insulator to Metal Transition, Spin-Phonon Coupling, and Potential Magnetic Transition Observed in Quantum Spin Liquid Candidate LiYbSe$_2$ under High Pressure

Authors: Haozhe Wang, Lifen Shi, Shuyuan Huyan, Greeshma C. Jose, Barbara Lavina, Sergey L. Bud'ko, Wenli Bi, Paul C. Canfield, Jinguang Cheng, Weiwei Xie

Abstract: Metallization of quantum spin liquid (QSL) materials has long been considered as a potential route to achieve unconventional superconductivity. Here we report our endeavor in this direction by pressurizing a three-dimensional QSL candidate, LiYbSe$_2$, with a previously unreported pyrochlore structure. High-pressure X-ray diffraction and Raman studies up to 50 GPa reveal no appreciable changes of… ▽ More Metallization of quantum spin liquid (QSL) materials has long been considered as a potential route to achieve unconventional superconductivity. Here we report our endeavor in this direction by pressurizing a three-dimensional QSL candidate, LiYbSe$_2$, with a previously unreported pyrochlore structure. High-pressure X-ray diffraction and Raman studies up to 50 GPa reveal no appreciable changes of structural symmetry or distortion in this pressure range. This compound is so insulating that its resistance decreases below 10$^5$ $Ω$ only at pressures above 25 GPa in the corresponding temperature range accompanying the gradual reduction of band gap upon compression. Interestingly, an insulator-to-metal transition takes place in LiYbSe$_2$ at about 68 GPa and the metallic behavior remains up to 123.5 GPa, the highest pressure reached in the present study. A possible sign of magnetic or other phase transition was observed in LiYbSe$_2$. The insulator-to-metal transition in LiYbSe$_2$ under high pressure makes it an ideal system to study the pressure effects on QSL candidates of spin-1/2 Yb$^{3+}$ system in different lattice patterns. △ Less

Submitted 4 January, 2024; v1 submitted 30 November, 2023; originally announced December 2023.

Strong enhancement of magnetic ordering temperature and structural/valence transitions in EuPd3S4 under high pressure

Authors: S. Huyan, D. H. Ryan, T. J. Slade, B. Lavina, G. C. Jose, H. Wang, J. M. Wilde, R. A. Ribeiro, J. Zhao, W. Xie, W. Bi, E. E. Alp, S. L. Bud'ko, P. C. Canfield

Abstract: We present a comprehensive study of the mixed valent compound, EuPd3S4, by electrical transport, X-ray diffraction, time-domain 151Eu synchrotron Mössbauer spectroscopy, and X-ray absorption spectroscopy measurements under high pressure. The electrical transport measurements show that the antiferromagnetic ordering temperature, TN, increases rapidly from 2.8 K at ambient pressure to 23.5 K at ~19… ▽ More We present a comprehensive study of the mixed valent compound, EuPd3S4, by electrical transport, X-ray diffraction, time-domain 151Eu synchrotron Mössbauer spectroscopy, and X-ray absorption spectroscopy measurements under high pressure. The electrical transport measurements show that the antiferromagnetic ordering temperature, TN, increases rapidly from 2.8 K at ambient pressure to 23.5 K at ~19 GPa and plateaus between ~19 and ~29 GPa after which no anomaly associated with TN is detected. A pressure-induced first order structural transition from cubic to tetragonal is observed, with a rather broad coexistence region (~20 GPa to ~32 GPa) that corresponds to the TN plateau. Mössbauer spectroscopy measurements show a clear valence transition from approximately 50:50 Eu2+:Eu3+ to fully Eu3+ at ~28 GPa, consistent with the vanishing of the magnetic order at the same pressure. X-ray absorption data show a transition to a fully trivalent state at a similar pressure. Our results show that pressure first greatly enhances TN, most likely via enhanced hybridization between the Eu 4f states and the conduction band, and then, second, causes a structural phase transition that coincides with the conversion of the europium to a fully trivalent state. △ Less

Submitted 28 June, 2023; originally announced June 2023.

Comments: 28 pages, 6 figures in main manuscript, 10 figures in SI

Journal ref: Proc. Natl. Acad. Sci. 120 (52) e2310779120 (2023)

Absolute Concentration Robustness in Rank-One Kinetic Systems

Authors: Eduardo R. Mendoza, Dylan Antonio SJ. Talabis, Editha C. Jose, Lauro L. Fontanil

Abstract: A kinetic system has an absolute concentration robustness (ACR) for a molecular species if its concentration remains the same in every positive steady state of the system. Just recently, a condition that sufficiently guarantees the existence of an ACR in a rank-one mass-action kinetic system was found. In this paper, it will be shown that this ACR criterion does not extend in general to power-law… ▽ More A kinetic system has an absolute concentration robustness (ACR) for a molecular species if its concentration remains the same in every positive steady state of the system. Just recently, a condition that sufficiently guarantees the existence of an ACR in a rank-one mass-action kinetic system was found. In this paper, it will be shown that this ACR criterion does not extend in general to power-law kinetic systems. Moreover, we also discussed in this paper a necessary condition for ACR in multistationary rank-one kinetic system which can be used in ACR analysis. Finally, a concept of equilibria variation for kinetic systems which are based on the number of the system's ACR species will be introduced here. △ Less

Submitted 7 April, 2023; originally announced April 2023.

Auxiliary Learning as a step towards Artificial General Intelligence

Authors: Christeen T. Jose

Abstract: Auxiliary Learning is a machine learning approach in which the model acknowledges the existence of objects that do not come under any of its learned categories.The name Auxiliary learning was chosen due to the introduction of an auxiliary class. The paper focuses on increasing the generality of existing narrow purpose neural networks and also highlights the need to handle unknown objects. The Cat… ▽ More Auxiliary Learning is a machine learning approach in which the model acknowledges the existence of objects that do not come under any of its learned categories.The name Auxiliary learning was chosen due to the introduction of an auxiliary class. The paper focuses on increasing the generality of existing narrow purpose neural networks and also highlights the need to handle unknown objects. The Cat & Dog binary classifier is taken as an example throughout the paper. △ Less

Submitted 30 November, 2022; originally announced December 2022.

Deepfake Detection using ImageNet models and Temporal Images of 468 Facial Landmarks

Authors: Christeen T Jose

Abstract: This paper presents our results and findings on the use of temporal images for deepfake detection. We modelled temporal relations that exist in the movement of 468 facial landmarks across frames of a given video as spatial relations by constructing an image (referred to as temporal image) using the pixel values at these facial landmarks. CNNs are capable of recognizing spatial relationships that e… ▽ More This paper presents our results and findings on the use of temporal images for deepfake detection. We modelled temporal relations that exist in the movement of 468 facial landmarks across frames of a given video as spatial relations by constructing an image (referred to as temporal image) using the pixel values at these facial landmarks. CNNs are capable of recognizing spatial relationships that exist between the pixels of a given image. 10 different ImageNet models were considered for the study. △ Less

Submitted 14 August, 2022; originally announced August 2022.

Wakeword Detection under Distribution Shifts

Authors: Sree Hari Krishnan Parthasarathi, Lu Zeng, Christin Jose, Joseph Wang

Abstract: We propose a novel approach for semi-supervised learning (SSL) designed to overcome distribution shifts between training and real-world data arising in the keyword spotting (KWS) task. Shifts from training data distribution are a key challenge for real-world KWS tasks: when a new model is deployed on device, the gating of the accepted data undergoes a shift in distribution, making the problem of t… ▽ More We propose a novel approach for semi-supervised learning (SSL) designed to overcome distribution shifts between training and real-world data arising in the keyword spotting (KWS) task. Shifts from training data distribution are a key challenge for real-world KWS tasks: when a new model is deployed on device, the gating of the accepted data undergoes a shift in distribution, making the problem of timely updates via subsequent deployments hard. Despite the shift, we assume that the marginal distributions on labels do not change. We utilize a modified teacher/student training framework, where labeled training data is augmented with unlabeled data. Note that the teacher does not have access to the new distribution as well. To train effectively with a mix of human and teacher labeled data, we develop a teacher labeling strategy based on confidence heuristics to reduce entropy on the label distribution from the teacher model; the data is then sampled to match the marginal distribution on the labels. Large scale experimental results show that a convolutional neural network (CNN) trained on far-field audio, and evaluated on far-field audio drawn from a different distribution, obtains a 14.3% relative improvement in false discovery rate (FDR) at equal false reject rate (FRR), while yielding a 5% improvement in FDR under no distribution shift. Under a more severe distribution shift from far-field to near-field audio with a smaller fully connected network (FCN) our approach achieves a 52% relative improvement in FDR at equal FRR, while yielding a 20% relative improvement in FDR on the original distribution. △ Less

Submitted 13 July, 2022; originally announced July 2022.

Latency Control for Keyword Spotting

Authors: Christin Jose, Joseph Wang, Grant P. Strimel, Mohammad Omar Khursheed, Yuriy Mishchenko, Brian Kulis

Abstract: Conversational agents commonly utilize keyword spotting (KWS) to initiate voice interaction with the user. For user experience and privacy considerations, existing approaches to KWS largely focus on accuracy, which can often come at the expense of introduced latency. To address this tradeoff, we propose a novel approach to control KWS model latency and which generalizes to any loss function withou… ▽ More Conversational agents commonly utilize keyword spotting (KWS) to initiate voice interaction with the user. For user experience and privacy considerations, existing approaches to KWS largely focus on accuracy, which can often come at the expense of introduced latency. To address this tradeoff, we propose a novel approach to control KWS model latency and which generalizes to any loss function without explicit knowledge of the keyword endpoint. Through a single, tunable hyperparameter, our approach enables one to balance detection latency and accuracy for the targeted application. Empirically, we show that our approach gives superior performance under latency constraints when compared to existing methods. Namely, we make a substantial 25\% relative false accepts improvement for a fixed latency target when compared to the baseline state-of-the-art. We also show that when our approach is used in conjunction with a max-pooling loss, we are able to improve relative false accepts by 25 % at a fixed latency when compared to cross entropy loss. △ Less

Submitted 14 June, 2022; originally announced June 2022.

Comments: Proceedings of INTERSPEECH

Complex balanced equilibria of weakly reversible poly-PL systems: existence, stability, and robustness

Authors: Editha C. Jose, Eduardo R. Mendoza, Dylan Antonio SJ. Talabis

Abstract: Poly-PL kinetic systems (PYK) are kinetic systems consisting of nonnegative linear combinations of power law functions. In this contribution, we analyze these kinetic systems using two main approaches: (1) we define a canonical power law representation of a poly-PL system, and (2) we transform a poly-PL system into a dynamically equivalent power law kinetic system that preserves the stoichiometric… ▽ More Poly-PL kinetic systems (PYK) are kinetic systems consisting of nonnegative linear combinations of power law functions. In this contribution, we analyze these kinetic systems using two main approaches: (1) we define a canonical power law representation of a poly-PL system, and (2) we transform a poly-PL system into a dynamically equivalent power law kinetic system that preserves the stoichiometric subspace of the system. These approaches led us to establish results that concern important dynamical properties of poly-PL systems that extend known results for generalized mass actions systems (GMAS) such as existence, uniqueness and parametrization of complex balanced steady states, and linear stability of complex balanced equilibria. Furthermore, the paper discusses subsets of poly-PL systems that exhibit two types of concentration robustness in some species namely absolute concentration robustness and balanced concentration robustness. △ Less

Submitted 23 December, 2021; originally announced December 2021.

Comments: 18 pages. arXiv admin note: substantial text overlap with arXiv:2006.10227

Tiny-CRNN: Streaming Wakeword Detection In A Low Footprint Setting

Authors: Mohammad Omar Khursheed, Christin Jose, Rajath Kumar, Gengshen Fu, Brian Kulis, Santosh Kumar Cheekatmalla

Abstract: In this work, we propose Tiny-CRNN (Tiny Convolutional Recurrent Neural Network) models applied to the problem of wakeword detection, and augment them with scaled dot product attention. We find that, compared to Convolutional Neural Network models, False Accepts in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using models based on the Tiny-CRNN architectu… ▽ More In this work, we propose Tiny-CRNN (Tiny Convolutional Recurrent Neural Network) models applied to the problem of wakeword detection, and augment them with scaled dot product attention. We find that, compared to Convolutional Neural Network models, False Accepts in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using models based on the Tiny-CRNN architecture, and we can get up to 32% reduction in False Accepts at a 50k parameter budget with 75% reduction in parameter size compared to word-level Dense Neural Network models. We discuss solutions to the challenging problem of performing inference on streaming audio with this architecture, as well as differences in start-end index errors and latency in comparison to CNN, DNN, and DNN-HMM models. △ Less

Submitted 29 September, 2021; originally announced September 2021.

Comments: arXiv admin note: substantial text overlap with arXiv:2011.12941

ACM Class: I.2.0

Absolutely complex balanced kinetic systems

Authors: Editha C. Jose, Eduardo R. Mendoza, Dylan Antonio SJ. Talabis

Abstract: A complex balanced kinetic system is absolutely complex balanced (ACB) if every positive equilibrium is complex balanced. Two results on absolute complex balancing were foundational for modern chemical reaction network theory (CRNT): in 1972, M. Feinberg proved that any deficiency zero complex balanced system is absolutely complex balanced. In the same year, F. Horn and R. Jackson showed that the… ▽ More A complex balanced kinetic system is absolutely complex balanced (ACB) if every positive equilibrium is complex balanced. Two results on absolute complex balancing were foundational for modern chemical reaction network theory (CRNT): in 1972, M. Feinberg proved that any deficiency zero complex balanced system is absolutely complex balanced. In the same year, F. Horn and R. Jackson showed that the (full) converse of the result is not true: any complex balanced mass action system, regardless of its deficiency, is absolutely complex balanced. In this paper, we present initial results on the extension of the Horn and Jackson ACB Theorem. In particular, we focus on other kinetic systems with positive deficiency where complex balancing implies absolute complex balancing. While doing so, we found out that complex balanced power law reactant determined kinetic systems (PL-RDK) systems are not ACB. In our search for necessary and sufficient conditions for complex balanced systems to be absolutely complex balanced, we came across the so-called CLP systems (complex balanced systems with a desired "log parametrization" property). It is shown that complex balanced systems with bi-LP property are absolutely complex balanced. For non-CLP systems, we discuss novel methods for finding sufficient conditions for ACB in kinetic systems containing non-CLP systems: decompositions, the Positive Function Factor (PFF) and the Coset Intersection Count (CIC) and their application to poly-PL and Hill-type systems. △ Less

Submitted 24 December, 2021; v1 submitted 20 March, 2021; originally announced March 2021.

Comments: 23 pages

MSC Class: 92C45; 65H10; 80A30

Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection

Authors: Mohammad Omar Khursheed, Christin Jose, Rajath Kumar, Gengshen Fu, Brian Kulis, Santosh Kumar Cheekatmalla

Abstract: In this work, we propose small footprint Convolutional Recurrent Neural Network models applied to the problem of wakeword detection and augment them with scaled dot product attention. We find that false accepts compared to Convolutional Neural Network models in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using CRNNs, and we can get up to 32% improvement… ▽ More In this work, we propose small footprint Convolutional Recurrent Neural Network models applied to the problem of wakeword detection and augment them with scaled dot product attention. We find that false accepts compared to Convolutional Neural Network models in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using CRNNs, and we can get up to 32% improvement at a 50k parameter budget with 75% reduction in parameter size compared to word-level Dense Neural Network models. We discuss solutions to the challenging problem of performing inference on streaming audio with CRNNs, as well as differences in start-end index errors and latency in comparison to CNN, DNN, and DNN-HMM models. △ Less

Submitted 25 November, 2020; originally announced November 2020.

Comments: \c{opyright} 2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works

Accurate Detection of Wake Word Start and End Using a CNN

Authors: Christin Jose, Yuriy Mishchenko, Thibaud Senechal, Anish Shah, Alex Escott, Shiv Vitaladevuni

Abstract: Small footprint embedded devices require keyword spotters (KWS) with small model size and detection latency for enabling voice assistants. Such a keyword is often referred to as \textit{wake word} as it is used to wake up voice assistant enabled devices. Together with wake word detection, accurate estimation of wake word endpoints (start and end) is an important task of KWS. In this paper, we prop… ▽ More Small footprint embedded devices require keyword spotters (KWS) with small model size and detection latency for enabling voice assistants. Such a keyword is often referred to as \textit{wake word} as it is used to wake up voice assistant enabled devices. Together with wake word detection, accurate estimation of wake word endpoints (start and end) is an important task of KWS. In this paper, we propose two new methods for detecting the endpoints of wake words in neural KWS that use single-stage word-level neural networks. Our results show that the new techniques give superior accuracy for detecting wake words' endpoints of up to 50 msec standard error versus human annotations, on par with the conventional Acoustic Model plus HMM forced alignment. To our knowledge, this is the first study of wake word endpoints detection methods for single-stage neural KWS. △ Less

Submitted 9 August, 2020; originally announced August 2020.

Comments: Proceedings of INTERSPEECH

Journal ref: Interspeech 2020

Complex balanced equilibria of weakly reversible poly-PL systems: multiplicity, robustness and stability

Authors: Noel T. Fortun, Dylan Antonio SJ. Talabis, Editha C. Jose, Eduardo R. Mendoza

Abstract: Poly-PL kinetic systems are kinetic systems consisting of nonnegative linear combinations of power law functions. In this contribution, we analyze these kinetic systems using two main approaches: (1) we define a canonical power law representation of a poly-PL system, and (2) we transform a poly-PL system into a dynamically equivalent power law kinetic system that preserves the stoichiometric subsp… ▽ More Poly-PL kinetic systems are kinetic systems consisting of nonnegative linear combinations of power law functions. In this contribution, we analyze these kinetic systems using two main approaches: (1) we define a canonical power law representation of a poly-PL system, and (2) we transform a poly-PL system into a dynamically equivalent power law kinetic system that preserves the stoichiometric subspace of the system. These approaches led us to establish results that concern important dynamical properties of poly-PL systems such as existence and parametrization of complex balanced steady states, capacity for multiple complex balanced equilibria in a stoichiometric compatibility class, concentration robustness of some species, and linear stability of complex balanced equilibria. △ Less

Submitted 12 July, 2021; v1 submitted 17 June, 2020; originally announced June 2020.

Comments: The uploader (Noel Fortun) dropped out of the project

Development of a Robust Depth-Pressure Estimation Algorithm for a Vision-Based Breast Self-Examination Guidance System

Authors: John Anthony C. Jose, Phoebe Mae L. Ching, Melvin K. Cabatuan

Abstract: In the case of breast cancer, as with most cancers, early detection can significantly improve a person's chances of survival. This makes it important for there to be an effective and accessible means of regularly checking for manifestations of the disease. A vision-based guidance system (VBGS) for breast self-examination (BSE) is one way to improve a person's ability to detect the cancerous system… ▽ More In the case of breast cancer, as with most cancers, early detection can significantly improve a person's chances of survival. This makes it important for there to be an effective and accessible means of regularly checking for manifestations of the disease. A vision-based guidance system (VBGS) for breast self-examination (BSE) is one way to improve a person's ability to detect the cancerous systems. In response to this need, this study sought to develop a depth-pressure estimation algorithm for the proposed VBGS. A large number of BSE videos were used to train the model, and these samples were segmented according to breast size, which was found to be a differentiation factor in the depth-pressure estimation. The result was an algorithm that was applicable for universal use. In addition to these, several feature extraction schemes were tested with the objective of making the algorithm functional on average technology. It was found that Law's Textures Histogram and Local Binary Pattern Global Histogram were the most effective in estimating pressure using visual data. Moreover, combinations of the two schemes further improved the accuracy of the model in estimation. The resulting algorithm was thereby fit to be used by the average consumer. △ Less

Submitted 27 August, 2019; originally announced August 2019.

The WILDTRACK Multi-Camera Person Dataset

Authors: Tatjana Chavdarova, Pierre Baqué, Stéphane Bouquet, Andrii Maksai, Cijo Jose, Louis Lettry, Pascal Fua, Luc Van Gool, François Fleuret

Abstract: People detection methods are highly sensitive to the perpetual occlusions among the targets. As multi-camera set-ups become more frequently encountered, joint exploitation of the across views information would allow for improved detection performances. We provide a large-scale HD dataset named WILDTRACK which finally makes advanced deep learning methods applicable to this problem. The seven-static… ▽ More People detection methods are highly sensitive to the perpetual occlusions among the targets. As multi-camera set-ups become more frequently encountered, joint exploitation of the across views information would allow for improved detection performances. We provide a large-scale HD dataset named WILDTRACK which finally makes advanced deep learning methods applicable to this problem. The seven-static-camera set-up captures realistic and challenging scenarios of walking people. Notably, its camera calibration with jointly high-precision projection widens the range of algorithms which may make use of this dataset. In aim to help accelerate the research on automatic camera calibration, such annotations also accompany this dataset. Furthermore, the rich-in-appearance visual context of the pedestrian class makes this dataset attractive for monocular pedestrian detection as well, since: the HD cameras are placed relatively close to the people, and the size of the dataset further increases seven-fold. In summary, we overview existing multi-camera datasets and detection methods, enumerate details of our dataset, and we benchmark multi-camera state of the art detectors on this new dataset. △ Less

Submitted 28 July, 2017; originally announced July 2017.

Kronecker Recurrent Units

Authors: Cijo Jose, Moustpaha Cisse, Francois Fleuret

Abstract: Our work addresses two important issues with recurrent neural networks: (1) they are over-parameterized, and (2) the recurrence matrix is ill-conditioned. The former increases the sample complexity of learning and the training time. The latter causes the vanishing and exploding gradient problem. We present a flexible recurrent neural network model called Kronecker Recurrent Units (KRU). KRU achiev… ▽ More Our work addresses two important issues with recurrent neural networks: (1) they are over-parameterized, and (2) the recurrence matrix is ill-conditioned. The former increases the sample complexity of learning and the training time. The latter causes the vanishing and exploding gradient problem. We present a flexible recurrent neural network model called Kronecker Recurrent Units (KRU). KRU achieves parameter efficiency in RNNs through a Kronecker factored recurrent matrix. It overcomes the ill-conditioning of the recurrent matrix by enforcing soft unitary constraints on the factors. Thanks to the small dimensionality of the factors, maintaining these constraints is computationally efficient. Our experimental results on seven standard data-sets reveal that KRU can reduce the number of parameters by three orders of magnitude in the recurrent weight matrix compared to the existing recurrent models, without trading the statistical performance. These results in particular show that while there are advantages in having a high dimensional recurrent space, the capacity of the recurrent part of the model can be dramatically reduced. △ Less

Submitted 31 December, 2017; v1 submitted 29 May, 2017; originally announced May 2017.

Understanding the non-linear clustering of high redshift galaxies

Authors: Charles Jose, Carlton M. Baugh, Cedric G. Lacey, Kandaswamy Subramanian

Abstract: We incorporate the non-linear clustering of dark matter halos, as modelled by Jose et al. (2016) into the halo model to better understand the clustering of Lyman break galaxies (LBGs) in the redshift range $z=3-5$. We find that, with this change, the predicted LBG clustering increases significantly on quasi-linear scales ($0.1 \leq r\,/\,h^{-1} \,{\rm Mpc} \leq 10$) compared to that in the linear… ▽ More We incorporate the non-linear clustering of dark matter halos, as modelled by Jose et al. (2016) into the halo model to better understand the clustering of Lyman break galaxies (LBGs) in the redshift range $z=3-5$. We find that, with this change, the predicted LBG clustering increases significantly on quasi-linear scales ($0.1 \leq r\,/\,h^{-1} \,{\rm Mpc} \leq 10$) compared to that in the linear halo bias model. This in turn results in an increase in the clustering of LBGs by an order of magnitude on angular scales $5" \leq θ\leq 100"$. Remarkably, the predictions of our new model on the whole remove the systematic discrepancy between the linear halo bias predictions and the observations. The correlation length and large scale galaxy bias of LBGs are found to be significantly higher in the non-linear halo bias model than in the linear halo bias model. The resulting two-point correlation function retains an approximate power-law form in contrast with that computed using the linear halo bias theory. We also find that the non-linear clustering of LBGs increases with increasing luminosity and redshift. Our work emphasizes the importance of using non-linear halo bias in order to model the clustering of high-z galaxies to probe the physics of galaxy formation and extract cosmological parameters reliably. △ Less

Submitted 6 June, 2017; v1 submitted 2 February, 2017; originally announced February 2017.

Comments: Accepted for publication in MNRAS

Scalable Metric Learning via Weighted Approximate Rank Component Analysis

Authors: Cijo Jose, Francois Fleuret

Abstract: We are interested in the large-scale learning of Mahalanobis distances, with a particular focus on person re-identification. We propose a metric learning formulation called Weighted Approximate Rank Component Analysis (WARCA). WARCA optimizes the precision at top ranks by combining the WARP loss with a regularizer that favors orthonormal linear mappings, and avoids rank-deficient embeddings. Usi… ▽ More We are interested in the large-scale learning of Mahalanobis distances, with a particular focus on person re-identification. We propose a metric learning formulation called Weighted Approximate Rank Component Analysis (WARCA). WARCA optimizes the precision at top ranks by combining the WARP loss with a regularizer that favors orthonormal linear mappings, and avoids rank-deficient embeddings. Using this new regularizer allows us to adapt the large-scale WSABIE procedure and to leverage the Adam stochastic optimization algorithm, which results in an algorithm that scales gracefully to very large data-sets. Also, we derive a kernelized version which allows to take advantage of state-of-the-art features for re-identification when data-set size permits kernel computation. Benchmarks on recent and standard re-identification data-sets show that our method beats existing state-of-the-art techniques both in term of accuracy and speed. We also provide experimental analysis to shade lights on the properties of the regularizer we use, and how it improves performance. △ Less

Submitted 23 March, 2016; v1 submitted 1 March, 2016; originally announced March 2016.

The clustering of dark matter halos: scale-dependent bias on quasi-linear scales

Authors: Charles Jose, Cedric G. Lacey, Carlton M. Baugh

Abstract: We investigate the spatial clustering of dark matter halos, collapsing from $1-4 σ$ fluctuations, in the redshift range $0 - 5$ using N-body simulations. The halo bias of high redshift halos ($z \geq 2$) is found to be strongly non-linear and scale-dependent on quasi-linear scales that are larger than their virial radii ($0.5-10$ Mpc/h). However, at lower redshifts, the scale-dependence of non-lin… ▽ More We investigate the spatial clustering of dark matter halos, collapsing from $1-4 σ$ fluctuations, in the redshift range $0 - 5$ using N-body simulations. The halo bias of high redshift halos ($z \geq 2$) is found to be strongly non-linear and scale-dependent on quasi-linear scales that are larger than their virial radii ($0.5-10$ Mpc/h). However, at lower redshifts, the scale-dependence of non-linear bias is weaker and and is of the order of a few percent on quasi-linear scales at $z \sim 0$. We find that the redshift evolution of the scale dependent bias of dark matter halos can be expressed as a function of four physical parameters: the peak height of halos, the non-linear matter correlation function at the scale of interest, an effective power law index of the {\it rms} linear density fluctuations and the matter density of the universe at the given redshift. This suggests that the scale-dependence of halo bias is not a universal function of the dark matter power spectrum, which is commonly assumed. We provide a fitting function for the scale dependent halo bias as a function of these four parameters. Our fit reproduces the simulation results to an accuracy of better than 4 % over the redshift range $0\leq z \leq 5$. We also extend our model by expressing the non-linear bias as a function of the linear matter correlation function. It is important to incorporate our results into the clustering models of dark matter halos at any redshift, including those hosting early generations of stars and galaxies before reionization. △ Less

Submitted 18 July, 2016; v1 submitted 22 September, 2015; originally announced September 2015.

Comments: Accepted for publication in MNRAS

Text Classification For Authorship Attribution Analysis

Authors: M. Sudheep Elayidom, Chinchu Jose, Anitta Puthussery, Neenu K Sasi

Abstract: Authorship attribution mainly deals with undecided authorship of literary texts. Authorship attribution is useful in resolving issues like uncertain authorship, recognize authorship of unknown texts, spot plagiarism so on. Statistical methods can be used to set apart the approach of an author numerically. The basic methodologies that are made use in computational stylometry are word length, senten… ▽ More Authorship attribution mainly deals with undecided authorship of literary texts. Authorship attribution is useful in resolving issues like uncertain authorship, recognize authorship of unknown texts, spot plagiarism so on. Statistical methods can be used to set apart the approach of an author numerically. The basic methodologies that are made use in computational stylometry are word length, sentence length, vocabulary affluence, frequencies etc. Each author has an inborn style of writing, which is particular to himself. Statistical quantitative techniques can be used to differentiate the approach of an author in a numerical way. The problem can be broken down into three sub problems as author identification, author characterization and similarity detection. The steps involved are pre-processing, extracting features, classification and author identification. For this different classifiers can be used. Here fuzzy learning classifier and SVM are used. After author identification the SVM was found to have more accuracy than Fuzzy classifier. Later combined the classifiers to obtain a better accuracy when compared to individual SVM and fuzzy classifier. △ Less

Submitted 18 October, 2013; originally announced October 2013.

Comments: 10 pages

Journal ref: Advanced Computing: An International Journal (ACIJ), Vol.4, No.5, September 2013

A physical model for the redshift evolution of high-z Lyman-Break Galaxies

Authors: Charles Jose, Raghunathan Srianand, Kandaswamy Subramanian

Abstract: We present a galaxy formation model to understand the evolution of stellar mass (M*) - UV luminosity relations, stellar mass functions and specific star formation rate (sSFR) of Lyman Break Galaxies (LBGs) along with their UV luminosity functions in the redshift range 3 < z < 8. Our models assume a physically motivated form for star formation in galaxies and model parameters are calibrated by fitt… ▽ More We present a galaxy formation model to understand the evolution of stellar mass (M*) - UV luminosity relations, stellar mass functions and specific star formation rate (sSFR) of Lyman Break Galaxies (LBGs) along with their UV luminosity functions in the redshift range 3 < z < 8. Our models assume a physically motivated form for star formation in galaxies and model parameters are calibrated by fitting the observed UV luminosity functions (LFs) of LBGs. We find the fraction of baryons that gets converted into stars remains nearly constant for z < 4 but shows an increase for z < 4. However, the rate of converting baryons into stars does not evolve significantly in the redshift range 3 < z < 8. Our model further successfully explains the M* - UV luminosity (M_AB) correlations of LBGs. While our model predictions of stellar mass functions compare well with the inferred data from observations at the low mass end, we need to invoke the Eddington bias to fit the high mass end. At any given redshift, we find the sSFR to be constant over the stellar mass range 5 \times 10^8 -5 \times 10^9 M_\odot and the redshift evolution of sSFR is well approximated by a form (1+z)^2.4 for 3 < z < 8 which is consistent with observations. Thus we find that dark matter halo build up in the LCDM model is sufficient to explain the evolution of UV LFs of LBGs along with their M* - M_AB relations, the stellar mass functions and the sSFR for 3 < z < 8. △ Less

Submitted 10 July, 2014; v1 submitted 15 October, 2013; originally announced October 2013.

Comments: 12 pages, 8 figures, Accepted for publication in MNRAS

Clustering at high redshift: The connection between Lyman Alpha emitters and Lyman break galaxies

Authors: Charles Jose, Raghunathan Srianand, Kandaswamy Subramanian

Abstract: We present a physically motivated semi-analytic model to understand the clustering of high redshift Lyman Alpha Emitters (LAEs). We show that the model parameters constrained by the observed luminosity functions, can be used to predict large scale bias and angular correlation function of LAEs. These predictions are shown to reproduce the observations remarkably well. We find that average masses of… ▽ More We present a physically motivated semi-analytic model to understand the clustering of high redshift Lyman Alpha Emitters (LAEs). We show that the model parameters constrained by the observed luminosity functions, can be used to predict large scale bias and angular correlation function of LAEs. These predictions are shown to reproduce the observations remarkably well. We find that average masses of dark matter halos hosting LAEs brighter than threshold narrow band magnitude ~ 25 are ~ 10^11 M_\odot. These are smaller than that of typical Lyman Break Galaxies (LBGs) brighter than similar threshold continuum magnitude by a factor ~ 10. This results in a smaller clustering strength of LAEs compared to LBGs. However, using the observed relationship between UV continuum and Lyman-alpha luminosity of LAEs, we show that both LAEs and LBGs belong to the same parent galaxy population with narrow band techniques having greater efficiency in picking up galaxies with low UV luminosity. We also show that the lack of evidence for the presence of the one halo term in the observed LAE angular correlation functions can be attributed to sub-Poisson distribution of LAEs in dark matter halos as a result of their low halo occupations. △ Less

Submitted 11 September, 2013; v1 submitted 28 April, 2013; originally announced April 2013.

Comments: Accepted for publication in MNRAS. arXiv admin note: text overlap with arXiv:1208.2097

Spatial Clustering of High Redshift Lyman Break Galaxies

Authors: Charles Jose, Kandaswamy Subramanian, Raghunathan Srianand, Saumyadip Samui

Abstract: We present a physically motivated semi-analytic model to understand the clustering of high redshift LBGs. We show that the model parameters constrained by the observed luminosity function, can be used to predict large scale (θ> 80 arcsec) bias and angular correlation function of galaxies. These predictions are shown to reproduce the observations remarkably well. We then adopt these model parameter… ▽ More We present a physically motivated semi-analytic model to understand the clustering of high redshift LBGs. We show that the model parameters constrained by the observed luminosity function, can be used to predict large scale (θ> 80 arcsec) bias and angular correlation function of galaxies. These predictions are shown to reproduce the observations remarkably well. We then adopt these model parameters to calculate the halo occupation distribution (HOD) using the conditional mass function. The halo model using this HOD is shown to provide a reasonably good fit to the observed clustering of LBGs at both large (θ>80") and small (θ< 10") angular scales for z=3-5 and several limiting magnitudes. However, our models underpredict the clustering amplitude at intermediate angular scales, where quasi-linear effects are important. The average mass of halos contributing to the observed clustering is found to be 6.2 x 10^{11} M_\odot and the characteristic mass of a parent halo hosting satellite galaxies is 1.2 \times 10^{12} M_\odot for a limiting absolute magnitude of -20.5 at z=4. For a given threshold luminosity these masses decrease with increasing z and at any given z these are found to increase with increasing value of threshold luminosity. We find that approximately 40 % of the halos above a minimum mass M_{min}, can host detectable central galaxies and about 5-10 % of these halos are likely to also host a detectable satellite. The satellites form typically a dynamical timescale prior to the formation of the parent halo. The small angular scale clustering is due to central-satellite pairs and is quite sensitive to changes in the duration of star formation in a halo. The present data favor star formation in a halo lasting typically for a few dynamical time-scales. Our models also reproduce different known trends between parameters related to star formation. △ Less

Submitted 29 November, 2012; v1 submitted 10 August, 2012; originally announced August 2012.

Comments: Accepted for publication in MNRAS

Weighing neutrinos using high redshift galaxy luminosity functions

Authors: Charles Jose, Saumyadip Samui, Kandaswamy Subramanian, Raghunathan Srianand

Abstract: Laboratory experiments measuring neutrino oscillations, indicate small mass differences between different mass eigenstates of neutrinos. The absolute mass scale is however not determined, with at present the strongest upper limits coming from astronomical observations rather than terrestrial experiments. The presence of massive neutrinos suppresses the growth of perturbations below a characteristi… ▽ More Laboratory experiments measuring neutrino oscillations, indicate small mass differences between different mass eigenstates of neutrinos. The absolute mass scale is however not determined, with at present the strongest upper limits coming from astronomical observations rather than terrestrial experiments. The presence of massive neutrinos suppresses the growth of perturbations below a characteristic mass scale, thereby leading to a decreased abundance of collapsed dark matter halos. Here we show that this effect can significantly alter the predicted luminosity function (LF) of high redshift galaxies. In particular we demonstrate that a stringent constraint on the neutrino mass can be obtained using the well measured galaxy LF and our semi-analytic structure formation models. Combining the constraints from the Wilkinson Microwave Anisotropy Probe 7 year (WMAP7) data with the LF data at z = 4, we get a limit on the sum of the masses of 3 degenerate neutrinos Σm_ν< 0.52 eV at the 95 % CL. The additional constraints using the prior on Hubble constant strengthens this limit to Σm_ν< 0.29 eV at the 95 % CL. This neutrino mass limit is a factor of order 4 improvement compared to the constraint based on the WMAP7 data alone, and as stringent as known limits based on other astronomical observations. As different astronomical measurements may suffer from different set of biases, the method presented here provides a complementary probe of Σm_ν. We suggest that repeating this exercise on well measured luminosity functions over different redshift ranges can provide independent and tighter constraints on Σm_ν. △ Less

Submitted 19 April, 2011; originally announced April 2011.

Comments: 14 pages, 7 figures, submitted to PRD

Journal ref: Phys.Rev.D83:123518,2011