Posters

New 2sED measure effectively bounds deep learning model complexity, correlating well with training error and offering efficient computation, particularly for deep models via a layerwise approach.

Trifle: Tractable inference for Offline RL achieves state-of-the-art results by using tractable generative models to overcome the inference-time suboptimality of existing sequence modeling approaches.

TACO, a novel topology-aware graph coarsening framework, tackles catastrophic forgetting in continual graph learning by efficiently preserving topological information during experience replay, signifi…

This paper fully characterizes concept classes optimistically universally learnable online, introducing novel algorithms and revealing equivalences between agnostic and realizable settings.

LLMs improve through self-correction, but the mechanisms are unclear. This paper provides a theoretical framework and empirical evidence demonstrating that self-correction arises from in-context align…

This paper theoretically analyzes speculative decoding, revealing its optimality and providing formulas for expected rejections, paving the way for more efficient large language model inference.

Boosting concept bottleneck model predictability, this paper introduces a theoretical framework linking concept set properties to model performance, proposing a method for effective concept identifica…

Bounded human feedback hinders large AI model training. This paper introduces hierarchical reinforcement learning to enable scalable oversight, efficiently acquiring feedback and learning optimal poli…

A lightweight knowledge alignment module enables two pre-trained LLMs to mutually learn and improve clinical language representation, exceeding individual model performance on various downstream tasks…

FedSAK, a novel federated multi-task learning framework, flexibly handles data, model, and task heterogeneity using tensor trace norm to learn correlations among client models, achieving superior perf…

Simple rule-based base-class mining (BCM) significantly boosts generalized few-shot semantic segmentation (GFSS) performance, surpassing complex existing methods.

Learn optimal device placement for neural networks with HSDAG, a novel framework boosting inference speed by up to 58.2%!

CounterAnimal: a new dataset exposes CLIP’s reliance on spurious correlations, challenging its perceived robustness and highlighting the need for more comprehensive evaluation benchmarks in vision-lan…

Single-step Sharpness-Aware Minimization (S2-SAM) achieves efficient and accurate sparse training by approximating sharpness perturbation via prior gradient information, incurring zero extra cost and …

UniDC framework achieves universal depth completion across various sensors and scenes using minimal labeled data, leveraging a foundation model and hyperbolic embedding for enhanced generalization.

S²GCSL: A novel scalable Granger causal structural learning approach efficiently identifies root causes of telecommunication network alarms by leveraging a linear kernel and incorporating expert knowl…

This paper introduces Bias-Conditioned Self-Influence (BCSI) for precise bias-conflicting sample detection and model rectification, enhancing fairness in machine learning.

SimGRL: A novel framework boosts visual reinforcement learning’s generalization by mitigating imbalanced saliency and observational overfitting through a feature-level frame stack and shifted random o…

A novel universal online learning algorithm achieves optimal gradient-variation regret across diverse function curvatures, boasting efficiency with only one gradient query per round.

A new adaptive learning rate for FTRL achieves minimax regret of O(T²/³) in online learning, improving existing best-of-both-worlds algorithms for various hard problems.