Posters

Treeffuser: Accurate probabilistic predictions from tabular data using conditional diffusion models and gradient-boosted trees!

TAP: automated jailbreaking of black-box LLMs with high success rates, using fewer queries than previous methods.

This paper optimizes randomized smoothing, a crucial certified defense against adversarial attacks, by introducing novel statistical methods that drastically reduce the computational cost, leading to …

Trap-MID: Outsmarting model inversion attacks with cleverly placed ’trapdoors'!

TransVIP: groundbreaking speech-to-speech translation system preserving voice & isochrony, outperforming current state-of-the-art models!

This paper presents a novel BayesOpt framework that incorporates Markov Decision Processes to optimize black-box functions with transition constraints, overcoming limitations of traditional methods.

Efficient Multi-Task Learning (EMTAL) transforms pre-trained Vision Transformers into efficient multi-task learners by using a MoEfied LoRA structure, a Quality Retaining optimization, and a router fa…

MOHAWK: Distilling Transformers’ quadratic knowledge into faster subquadratic SSMs, achieving state-of-the-art performance with <1% of training data!

Transformers encode information beyond next-token prediction by linearly representing belief state geometry in their residual stream, even with complex fractal structures.

Large language models (LLMs) suffer from information loss due to representational collapse and over-squashing, causing failures in simple tasks; this paper provides theoretical analysis and practical …

Transformers surprisingly learn second-order optimization methods for in-context linear regression, achieving exponentially faster convergence than gradient descent!

Researchers enhanced transformer performance on arithmetic tasks by introducing Abacus Embeddings, which encode each digit’s position, enabling improved generalization and unlocking multi-step reasoni…

Pre-trained transformers can provably learn to play games near-optimally using in-context learning, offering theoretical guarantees for both decentralized and centralized settings.

Transformers excel at in-context learning by leveraging minimax-optimal nonparametric learning, achieving near-optimal risk with sufficient pretraining data diversity.

Transformer Doctor diagnoses and treats vision transformer errors by identifying and correcting information integration issues, improving model performance and interpretability.

This paper introduces a novel theoretical framework for robust machine learning under distribution shifts, offering learning rules and guarantees, highlighting the game-theoretic viewpoint of distribu…

This paper bridges the gap between synthetic and real image disentanglement by proposing a novel transfer learning approach. The method leverages weakly supervised learning on synthetic data to train…

Transferable Boltzmann Generators enable efficient, zero-shot sampling of unseen molecular systems’ equilibrium distributions, boosting molecular simulations.

UMI-GRAT: A universal meta-initialized and gradient robust adversarial attack effectively exploits vulnerabilities in the Segment Anything Model (SAM) and its fine-tuned downstream models, even withou…

Challenging common assumptions, researchers prove that flatter adversarial examples don’t guarantee better transferability and introduce TPA, a theoretically-grounded attack creating more transferable…