12 papers
Entropic Projection Alignment: Estimating, Explaining, and Improving Model Performance Under Distribution Shift
Salim I. Amoukou, Emanuele Albini, Tom Bewley +2
We propose a unified framework for addressing three key challenges of distribution shift: (1) estimating a model's performance on an unlabeled target domain, (2) explaining the shi…
Correcting Split Selection in Online Decision Trees via Anytime-Valid Inference
Salim I. Amoukou, Saumitra Mishra, Manuela Veloso
Bagging-based ensembles, most notably Adaptive Random Forests, are among the strongest performers for learning from data streams. A common denominator across these methods is their…
ShapShift: Explaining Model Prediction Shifts with Subgroup Conditional Shapley Values
Tom Bewley, Salim I. Amoukou, Emanuele Albini +2
Changes in input distribution can induce shifts in the average predictions of machine learning models. Such prediction shifts may impact downstream business outcomes (e.g. a bank's…
Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs
Ngoc Bui, Shubham Sharma, Simran Lamba +2
Memory and computation remain core bottlenecks in long-horizon LLM inference due to the quadratic cost of self-attention and the ever-growing key-value (KV) cache. Existing strateg…
Representation Consistency for Accurate and Coherent LLM Answer Aggregation
Junqi Jiang, Tom Bewley, Salim I. Amoukou +4
Test-time scaling improves large language models' (LLMs) performance by allocating more compute budget during inference. To achieve this, existing methods often require intricate m…
To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Models
Anna Hedström, Salim I. Amoukou, Tom Bewley +2
We introduce Mechanistic Error Reduction with Abstention (MERA), a principled framework for steering language models (LMs) to mitigate errors through selective, adaptive interventi…