🏢 Cornell University

WorldCoder: an LLM agent builds world models via code generation and interaction, proving highly sample-efficient and enabling knowledge transfer.

Recommendation systems must balance user satisfaction with fair item exposure. This research provides a theoretical model and empirical validation showing that user preference diversity can significan…

MPI: A novel framework harnesses multi-modal biological data via view decoupling for superior phenotype imputation.

Poisson Midpoint Method quadratically accelerates Langevin Monte Carlo for diffusion models, achieving high-quality image generation with significantly fewer computations.

This research dramatically accelerates and improves hybrid language models by distilling large Transformers into linear RNNs, achieving performance comparable to the original Transformer with signific…

This paper presents computationally efficient transfer reinforcement learning algorithms that remove the dependence on state/action space sizes while achieving minimax optimality.

Unlocking the mysteries of stochastic approximation with constant stepsize, this paper reveals how memory and nonlinearity interact to create bias, providing novel analysis and solutions for more accu…

Supervised Kernel Thinning accelerates kernel regression by cleverly compressing data, achieving quadratic speedups in training and inference with minimal accuracy loss.

This paper reveals how many buyer samples are needed to set near-optimal posted prices for a single item, resolving a fundamental problem in online markets and offering both theoretical and practical …

REBEL, a novel reinforcement learning algorithm, simplifies policy optimization by regressing relative rewards, achieving strong performance in language and image generation tasks with increased effic…

Researchers introduce QIM-compatibility, a novel framework for modeling qualitative relationships in probability distributions using directed hypergraphs, significantly expanding beyond standard condi…

QTIP: Ultra-high dimensional LLM quantization using trellis codes for faster, higher-quality inference.

Neural Gaffer: Relighting any object via diffusion using a single image and an environment map to produce high-quality, realistic relit images.

ChemFlow: a new framework efficiently explores chemical space using latent flows, unifying existing methods & incorporating physical priors for molecule manipulation and optimization.

This paper introduces an efficient multivariate stochastic dominance test using optimal transport, enabling robust model benchmarking by considering metric dependencies.

LLMs struggle with graph recall, exhibiting biases like favoring triangles and underperforming compared to humans; advanced models show striking domain dependence.

This paper proves that language generation in the limit is always possible, even with an adversarial setting, contrasting with the impossibility of language identification in the limit.

Large Language Models (LLMs) surprisingly improve the challenging task of Programming by Example (PBE) when fine-tuned on problem-specific data, outperforming classic symbolic methods and even surpass…

Researchers uniquely estimate probabilities in Causal Bayesian Networks using simple independence assumptions, enabling analysis from observational data and simplifying counterfactual probability calc…

Hybrid causal discovery algorithm efficiently learns unique causal graphs from observational data by leveraging local substructures and topological sorting, outperforming existing methods in accuracy …