Instruction tuning improves the ability of large language models (LLMs) to follow diverse human instructions, but achieving strong performance on specific target tasks remains challenging. A critical bottleneck is selecting the most relevant data to maximize task-specific performance. Existing data selection approaches include unstable influence-based methods and more stable distribution alignment methods, the latter of which critically rely on the underlying sample representation. In practice, most distribution alignment methods, from shallow features (e.g., BM25) to neural embeddings (e.g., BGE, LLM2Vec), may fail to capture how the model internally processes samples. To bridge this gap, we adopt a model-centric strategy in which each sample is represented by its neuronal activation pattern in the model, directly reflecting internal computation. However, directly using raw neuron activations leads to spurious similarity between unrelated samples due to neuron polysemanticity, where a single neuron may respond to multiple, unrelated concepts. To address this, we employ sparse autoencoders to disentangle polysemantic activations into sparse, monosemantic representations, and introduce a dedicated similarity metric for this space to better identify task-relevant data. Comprehensive experiments across multiple instruction datasets, models, tasks, and selection ratios show that our approach consistently outperforms existing data selection baselines in both stability and task-specific performance2.

Evaluating representation similarity is fundamental to representation learning. However, existing metrics suffer from significant limitations: they lack interpretability due to shifting baselines, lack robustness to outliers, and are computationally intractable for large datasets, forcing reliance on heuristic approximations. To address this, we develop an ordinal-similarity framework, instan-Figure 1. TSI and QSI measure alignment between two representiated by the Triplet (TSI) and Quadruplet (QSI) tation spaces (e.g., Visual and Textual) by quantifying the conSimilarity Indices, which measure alignment bysistency of ordinal relationships. TSI checks if relative similarity quantifying the consistency of ordinal relation-from an anchor is preserved (e.g., 'Is Acloser to B than to C?'). QSI compares relative similarity between distinct pairs (e.g., 'Is A ships. We theoretically demonstrate this formu-closer to B than C is to D?') lation is inherently interpretable, robust to outliers, and computationally efficient. Finally, wemodel design and behavioral analysis, the reliability of these establish a formal equivalence between TSI andmetrics is paramount for the interpretability of increasingly local neighborhood alignment, measured by Mu-ubiquitous AI systems.

Neural Information Processing Systems http://nips.cc/

Neural Information Processing Systems http://nips.cc/

We consider the problem of online learning in the linear contextual bandits setting, but in which there are also strong individual fairness constraints governed by an unknown similarity metric. These constraints demand that we select similar actions or individuals with approximately equal probability DHPRZ12, which may be at odds with optimizing reward, thus modeling settings where profit and social policy are in tension. We assume we learn about an unknown Mahalanobis similarity metric from only weak feedback that identifies fairness violations, but does not quantify their extent. This is intended to represent the interventions of a regulator who knows unfairness when he sees it but nevertheless cannot enunciate a quantitative fairness metric over individuals. Our main result is an algorithm in the adversarial context setting that has a number of fairness violations that depends only logarithmically on T, while obtaining an optimal O(sqrt(T)) regret bound to the best fair policy.

Neural Information Processing Systems http://nips.cc/

An important step to understanding and improving artificial vision systems is to measure image similarity purely based on intrinsic object properties that define object identity. This problem has been studied in the computer vision literature as re-identification, though mostly restricted to specific object categories such as people and cars. We propose to extend it to general object categories, exploring an image similarity metric based on object intrinsics.

Neural Information Processing Systems http://nips.cc/