🏢 University of Washington

Boosting vision-language models: Multilingual data improves performance on English-centric benchmarks.

Meta-DiffuB enhances sequence-to-sequence text diffusion models by using meta-exploration to learn a contextualized noise schedule, resulting in state-of-the-art performance.

MEDIQ benchmark revolutionizes LLM evaluation by shifting from static to interactive clinical reasoning, revealing LLMs’ struggles with proactive information-seeking and highlighting the importance of…

MAGNET, a novel adaptive gradient-based tokenization method, tackles multilingual language model bias by employing language-specific boundary predictors to achieve equitable segmentation across divers…

This paper develops efficient algorithms for pricing homogeneous data in online settings, achieving low regret using novel discretization schemes that scale well with data size and number of buyer typ…

Generative Agent Modeling for Multi-agent Adaptation (GAMMA) improves human-AI cooperation by training AI agents against diverse partners generated from a latent model, enhancing zero-shot coordinatio…

AI learns optimal taxes for congestion games, maximizing social welfare with limited feedback, via a novel algorithm.

TABULA-8B, a novel language model for tabular prediction, achieves state-of-the-art zero-shot and few-shot performance across various benchmarks, exceeding existing methods by 5-15 percentage points.

Zero Graves-Lai constant ensures both bounded regret and delay robustness in online decision-making, particularly for linear models.

Adaptively initializing multi-service ML systems for diverse users using minimal data, this paper introduces a randomized algorithm achieving near-optimal loss with provable guarantees.

Neural networks learn features effectively through gradient descent, not just at the beginning, but also at the end of training, even with carefully regularized objectives.

ODIN, trained on a million 360° videos (360-1M), generates realistic novel views and reconstructs 3D scenes from single images.

DRAGO: A novel primal-dual algorithm delivers faster, state-of-the-art convergence for distributionally robust optimization.

DiSPOs: a novel model for zero-shot policy optimization in reinforcement learning, enabling quick adaptation to new tasks by learning a distribution of successor features and avoiding compounding erro…

AI discovers robust, biologically-plausible plasticity rules that self-organize and maintain neural circuits’ sequential activity, even with synaptic turnover.

Deep Submodular Peripteral Networks (DSPNs) learn submodular functions efficiently using graded pairwise comparisons, surpassing traditional methods and demonstrating superiority in experimental desig…

Researchers uncover hidden training data secrets of large language models by analyzing their byte-pair encoding tokenizers, revealing the proportions of different languages and domains.

Multi-Sub leverages multi-modal learning to achieve customized multiple clustering, aligning user-defined textual preferences with visual representations via a subspace proxy learning framework.

Score estimation, crucial for diffusion models, is computationally hard even with polynomial sample complexity unless strong distributional assumptions are made.

Boosting multimodal contrastive learning, this research introduces negCLIPLoss and NormSim, novel data selection methods surpassing existing techniques by improving data quality and task relevance. Th…