Posters

ReEvo, a novel integration of evolutionary search and LLM reflections, generates state-of-the-art heuristics for combinatorial optimization problems, demonstrating superior sample efficiency.

REDUCR, a novel data downsampling method, significantly improves worst-class test accuracy in imbalanced datasets by using class priority reweighting, surpassing state-of-the-art methods by ~15%.

Cross-Layer Attention (CLA) shrinks Transformer Key-Value cache 2x, improving LLMs’ memory efficiency without accuracy loss.

RISE: Recursive Introspection teaches LLMs to iteratively improve their responses, enabling self-correction and enhanced performance on challenging reasoning tasks.

Memoroids and Tape-Based Batching revolutionize recurrent RL, enabling efficient processing of long sequences and improving sample efficiency by eliminating segmentation.

Recurrent neural networks struggle with long-term memory due to a newly identified ‘curse of memory’: increasing parameter sensitivity with longer memory. This work provides insights into RNN optimiza…

SynCx, a novel recurrent autoencoder with complex weights, surpasses state-of-the-art models in unsupervised object discovery by iteratively refining phase relationships to achieve robust object bindi…

RectifID personalizes image generation by cleverly guiding a diffusion model using off-the-shelf classifiers, achieving identity preservation without needing extra training data.

Boost skeleton action recognition accuracy across datasets by recovering complete actions and resampling; outperforms existing methods.

Researchers developed a novel method for reconstructing the 3D shape of manipulated garments, achieving superior accuracy compared to existing techniques, particularly for complex, non-rigid deformati…

Deleting data from machine learning models exposes individuals to highly accurate reconstruction attacks, even when models are simple; this research demonstrates the vulnerability.

SRStitcher revolutionizes image stitching by integrating fusion and rectangling into a unified inpainting model, eliminating model training and achieving superior performance and stability.

RegionSpot efficiently integrates pretrained localization and vision-language models for superior open-world object recognition, achieving significant performance gains with minimal training.

Reciprocators: AI agents that learn to cooperate by reciprocating influence, achieving prosocial outcomes in complex scenarios.

Numerous machine learning algorithms are unified under the novel paradigm of reciprocal learning, proven to converge at linear rates under specific conditions, enhancing sample efficiency.

REBORN: An iterative training framework significantly improves unsupervised ASR by learning optimal speech segment boundaries using reinforcement learning, outperforming existing methods.

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…

Overparameterized neural networks surprisingly recover from catastrophic interference when trained cyclically on repeated data sequences, exhibiting anticipatory knowledge reactivation.

Boosting large language model performance after knowledge editing: A new method (D4S) minimizes model damage by regulating the explosive growth of parameter layers, enabling multiple effective edits.

BeTopNet uses braid theory to create a topological representation of multi-agent future driving behaviors, improving prediction and planning accuracy in autonomous driving systems.