Posters

Siamese Transformer with Hierarchical Refinement achieves state-of-the-art lane detection accuracy by integrating global and local features and a novel Curve-IoU loss.

Stable oracles outperform Gaussian oracles in high-accuracy heavy-tailed sampling, overcoming limitations of Gaussian-based proximal samplers.

New scalable algorithm reconstructs brain dynamics from short neuroimaging data, overcoming limitations of existing methods and enabling more accurate, efficient analysis of large-scale brain activity…

This paper introduces a novel, robust inlier identification algorithm for point cloud registration that leverages lo-minimization.

A novel machine learning recipe drastically accelerates charge density prediction in density functional theory, achieving state-of-the-art accuracy while being significantly faster than existing metho…

New framework directly controls neural network sensitivity by precisely parameterizing overall bi-Lipschitzness, offering improved robustness and generalization.

KG-ICL, a novel prompt-based knowledge graph foundation model, achieves universal in-context reasoning by leveraging in-context learning and a unified tokenizer, outperforming various baselines on 43 …

A novel probability-based contrastive learning framework significantly improves 3D molecular representation learning by mitigating false pairs, achieving state-of-the-art results.

BLOCC, a novel first-order algorithm, efficiently solves bilevel optimization problems with coupled constraints, offering improved scalability and convergence for machine learning applications.

LLMs spontaneously encode syntax using a polar coordinate system, representing syntactic relations via relative direction and distance of word embeddings.

Deep learning models often suffer from overconfidence; this paper introduces a PID controller to adaptively adjust a probability-dependent gradient decay rate, ensuring consistent optimization of both…

GeCo: A novel single-stage low-shot counter achieving ~25% improvement in count accuracy, via unified object detection, segmentation, and counting.

Estimating heterogeneous treatment effects (CATE) under unmeasured confounding is revolutionized by a novel non-parametric direct learning approach using instrumental variables, offering efficient and…

Multi-task learning with shallow ReLU networks yields almost always unique solutions equivalent to kernel methods, unlike single-task settings.

MQL-UCB: Near-optimal reinforcement learning with low policy switching cost, solving the exploration-exploitation dilemma for complex models.

A novel motion-aware spatio-temporal graph model surpasses existing methods in video salient object ranking by jointly optimizing multi-scale spatial and temporal features, thus accurately prioritizin…

A novel modular diffusion framework for image reconstruction dramatically cuts computational costs and achieves state-of-the-art perceptual quality across various tasks by cleverly combining pre-train…

New empirical methodology quantifies how much reinforcement learning algorithm performance relies on per-environment hyperparameter tuning, enabling better algorithm design.

Metalearning a neural circuit mimics nonparametric Bayesian inference, enabling fast, accurate, open-set classification.

LNGD: A Layer-Wise Natural Gradient optimizer drastically cuts deep neural network training time without sacrificing accuracy.