https://avidml.org/taxonomy/Recent content in Taxonomy Library on AVIDHugo -- gohugo.ioen-ushttps://avidml.org/taxonomy/ethics/Mon, 01 Jan 0001 00:00:00 +0000https://avidml.org/taxonomy/ethics/This domain is intended to codify ethics-related, often unintentional failure modes, e.g. algorithmic bias, misinformation. ID Name Description E0100 Bias/Discrimination Concerns of algorithms propagating societal bias E0101 Group fairness Fairness towards specific groups of people E0102 Individual fairness Fairness in treating similar individuals E0200 Explainability Ability to explain decisions made by AI E0201 Global explanations Explain overall functionality E0202 Local explanations Explain specific decisions E0300 User actions Perpetuating/causing/being affected by negative user actions E0301 Toxicity Users hostile towards other users E0302 Polarization/ Exclusion User behavior skewed in a significant direction E0400 Misinformation Perpetuating/causing the spread of falsehoods E0401 Deliberative Misinformation Generated by individuals.https://avidml.org/taxonomy/performance/Mon, 01 Jan 0001 00:00:00 +0000https://avidml.org/taxonomy/performance/This domain is intended to codify deficiencies such as privacy leakage or lack or robustness. ID Name Description P0100 Data issues Problems arising due to faults in the data pipeline P0101 Data drift Input feature distribution has drifted P0102 Concept drift Output feature/label distribution has drifted P0103 Data entanglement Cases of spurious correlation and proxy features P0104 Data quality issues Missing or low-quality features in data P0105 Feedback loops Unaccounted for effects of an AI affecting future data collection P0200 Model issues Ability for the AI to perform as intended P0201 Resilience/stability Ability for outputs to not be affected by small change in inputs P0202 OOD generalization Test performance doesn’t deteriorate on unseen data in training P0203 Scaling Training and inference can scale to high data volumes P0204 Accuracy Model performance accurately reflects realistic expectations P0300 Privacy Protect leakage of user information as required by rules and regulations P0301 Anonymization Protects through anonymizing user identity P0302 Randomization Protects by injecting noise in data, eg.https://avidml.org/taxonomy/security/Mon, 01 Jan 0001 00:00:00 +0000https://avidml.org/taxonomy/security/This domain is intended to codify the landscape of threats to a ML system. ID Name Description S0100 Software Vulnerability Vulnerability in system around model—a traditional vulnerability S0200 Supply Chain Compromise Compromising development components of a ML model, e.g. data, model, hardware, and software stack. S0201 Model Compromise Infected model file S0202 Software compromise Upstream Dependency Compromise S0300 Over-permissive API Unintended information leakage through API S0301 Information Leak Cloud Model API leaks more information than it needs to S0302 Excessive Queries Cloud Model API isn’t sufficiently rate limited S0400 Model Bypass Intentionally try to make a model perform poorly S0401 Bad Features The model uses features that are easily gamed by the attacker S0402 Insufficient Training Data The bypass is not represented in the training data S0403 Adversarial Example Input data points intentionally supplied to draw mispredictions.