🏢 UC Berkeley

This research introduces the first theoretical guarantees for warm-starting the celebrated Push-Relabel network flow algorithm, improving its speed using a predicted flow, while maintaining worst-case…

LLMLIFT: An LLM-powered approach builds verified lifting tools for DSLs, outperforming prior symbolic methods in benchmark transpilation and requiring less development effort.

Boosting clinical trial efficiency, this research introduces a covariate-adjusted response-adaptive randomization (CARA) design that effectively leverages surrogate outcomes to handle delayed primary …

A revolutionary ultrafast phylogenetic method outperforms protein language models in variant effect prediction by efficiently estimating amino acid substitution rates from massive datasets.

Researchers developed Subsampled Smooth Calibration Error (SSCE), a new truthful calibration measure for sequential prediction, solving the problem of existing measures being easily gamed.

LLMs struggle with simple logical reasoning due to the ‘reversal curse.’ This paper reveals that weight asymmetry during training is the culprit, offering a new theoretical perspective and potential s…

ESCAIP, a novel neural network architecture, dramatically boosts the speed and accuracy of atomic simulations by leveraging attention mechanisms, enabling efficient large-scale modeling across diverse…

LoRA’s initialization significantly impacts finetuning; initializing matrix A randomly and B to zero yields better performance than vice-versa due to enabling larger learning rates.

LLMs struggle with very low-resource programming languages. SPEAC, a novel synthetic programming elicitation and compilation approach, uses an intermediate language to enable LLMs to generate syntact…

SSDM: Scalable Speech Dysfluency Modeling tackles challenges in speech dysfluency analysis by using articulatory gestures for scalable alignment, a connectionist subsequence aligner for efficient dysf…

SGLang: A new system boosts LLM program execution speed by up to 6.4x, simplifying complex LLM application programming.

Unsupervised SAM (UnSAM) achieves competitive image segmentation results without human annotation, surpassing previous unsupervised methods and even improving supervised SAM’s accuracy.

AI agents can secretly collude using steganography, hiding their interactions from oversight. This research formalizes this threat, analyzes LLMs’ capabilities, and proposes mitigation strategies.

Deep learning’s neural scaling laws defy conventional wisdom; this paper uses infinite-dimensional linear regression to theoretically explain this phenomenon, showing that implicit regularization of S…

Researchers enhanced image generation by improving score distillation sampling via a novel Schrödinger Bridge framework, improving realism without computational overhead.

This paper reveals the surprising computational hardness of achieving fairness in low-rank approximation while offering efficient approximation algorithms.

New algorithms conquer regret in multi-armed bandits challenged by network interference, achieving provably low regret with both known and unknown network structures.

New metric, λ-discrepancy, precisely detects & mitigates partial observability in sequential decision processes, significantly boosting reinforcement learning agent performance.

Med-Real2Sim uses physics-informed self-supervised learning to build non-invasive medical digital twins, enabling in-silico clinical trials and unsupervised disease detection.

Unlocking complex models’ secrets: New algorithm identifies input interactions using the Möbius Transform, boosting interpretability with surprising speed and accuracy.