Localising Shortcut Learning in Pixel Space via Ordinal Scoring Correlations for Attribution Representations (OSCAR) Deep neural networks often exploit shortcuts. These are spurious cues which are associated with output labels in the training data but are unrelated to task semantics. When the shortcut features are a...

📄 OSCAR: an auditing framework for localising shortcut reliance in pixel space.

We turn attribution maps into interpretable statistics to quantify spurious effects. Feedback welcome!
arxiv.org/abs/2512.18888
#ComputerVision #FairML #ExplainableAI #XAI #AI

Analyzing Dialectical Biases in LLMs for Knowledge and Reasoning Benchmarks Large language models (LLMs) are ubiquitous in modern day natural language processing. However, previous work has shown degraded LLM performance for under-represented English dialects. We analyze the ...

Please read our actual paper for more interesting details! 📄 arxiv.org/abs/2510.00962
Come discuss more about #FairML at our presentation session! Happening (right now!!!) on Nov 5th 7PM EST at Gather Session 3!

Fair Set-Valued Classification with Demographic Parity Constraints

Researchers propose oracle and proxy methods that enforce demographic parity in set-valued classifiers while keeping label-set size near target; proxy matches oracle fairness with lower runtime. getnews.me/fair-set-valued-classifi... #fairml #setvalued

Overgeneralization in Fair Machine Learning: Why Identities Matter

The study presented at ACM FAccT 2025 warns that using interchangeable fairness metrics for race, gender, ability, and age can miss unique harms; the preprint debuted on May 7 2025. Read more: getnews.me/overgeneralization-in-fa... #fairml #facc2025

TABFAIRGDT: Fast Fair Synthetic Tabular Data via Autoregressive Decision Trees

TABFAIRGDT achieves a 72% speedup over DL baselines and can synthesize 10 k rows in about one second on a standard CPU. Accepted for IEEE ICDM 2025. Read more: getnews.me/tabfairgdt-fast-fair-syn... #tabfairgdt #fairml #syntheticdata

Optimal Steering Method Guarantees Exact Fairness in AI Models

Researchers unveiled an optimal steering technique that uses KL divergence to reshape data toward an ideal distribution, achieving exact demographic parity on the Bios dataset. Read more: getnews.me/optimal-steering-method-... #fairml #optimalsteering #ai

Addressing Group‑Specific Concept Drift for Fair Federated Learning

New federated‑learning framework spots group‑specific concept drift using local detectors and model updates, cutting fairness violations on benchmark tests while keeping accuracy. getnews.me/addressing-group-specifi... #federatedlearning #fairml

When is Automated Decision Making Legitimate? A book chapter summary by Esha Srivastava

Chapter 2 of Fairness & ML strongly emphasizes on the need for justifiable and interpretable decisions — not just accurate ones.

𝐁𝐥𝐨𝐠 𝐋𝐢𝐧𝐤: aiineverything.blogspot.com/2025/07/when...

@tsonika.bsky.social @tyagilab.bsky.social

#AIethics #FairML #ResponsibleAI #DigitalHealthReadingGroup #TyagiLab

Ok #fairml & algo-fairness people, what are good first sentences for papers that end this madness of "ml is more and more used to make consequential decisions"???

Kudos to the incredible CDI team: Sejin Kim, Joshua Siraj, Muammar Kabir, Mattea Welch, Clare McElcheran, Tran Truong👏This is a big step toward scalable, robust, and #fairML frameworks in #oncology

@pmresearch-uhn.bsky.social @uhnresearch.bsky.social @uhntoronto.bsky.social @uhnaihub.bsky.social

Kudos to the incredible CDI team: Sejin Kim, Joshua Siraj, Muammar Kabir, Mattea Welch, Clare McElcheran, Tran Truong👏This is a big step toward scalable, robust, and #fairML frameworks in #oncology

@pmresearch-uhn.bsky.social @uhnresearch.bsky.social @uhntoronto.bsky.social @uhnaihub.bsky.social