Procedural Fairness in Multi-Agent Bandits

Research introduces procedural fairness, defined as equal voice and representation within a policy, for multi-agent multi-armed bandits.

Research explores integrating causality into algorithmic recourse to ensure recommended changes genuinely improve qualifications, not just game classifiers.

LightRot introduces a lightweight rotation scheme and hardware architecture for energy-efficient, accurate low-bit large language model inference.

ReToken introduces a single learnable embedding to improve visual retrieval in vision-language models by selecting relevant visual tokens, addressing long visual context challenges.

GyRot proposes a quantization framework and hardware accelerator to improve low-bit LLM inference by integrating rotation and fine-grained group quantization.

Research proposes Quadratic Objective Perturbation (QOP) for differentially private empirical risk minimization, addressing limitations of Linear Objective Perturbation (LOP) with unbounded gradients.

Research proposes Neuromorphic Diffusion Language Models to reduce LLM inference compute and memory bottlenecks through sparsity and block denoising.

CLBench-V introduces a new benchmark for evaluating multimodal context learning, focusing on models' ability to learn from diverse contexts beyond text.

Researchers propose Addressable Recall Compaction (ARC), a new context-management framework for LLM agents to overcome context window limits by separating archival storage from active context.

New research introduces Stemma, a method to determine LLM provenance by mapping 'induced decision regions,' improving reliability over response-level analysis.

Research paper explores how Large Language Models internally collapse reading and writing into a single entangled autoregressive process, unlike human brains.

Research identifies methods to detect knowledge inconsistencies across multimodal data (text, tables, knowledge graphs) from sources like Wikipedia and Wikidata.