Posters

This paper presents novel, privacy-preserving algorithms achieving near-optimal rates for solving stochastic saddle point problems and variational inequalities in non-Euclidean geometries.

Train high-quality generative models with strong differential privacy using a novel slicing mechanism that injects noise into random low-dimensional data projections, avoiding noisy gradients.

Researchers reveal ‘privacy backdoors,’ a new attack that exploits pre-trained models to leak user training data, highlighting critical vulnerabilities and prompting stricter model security measures.

Prism: a novel framework disentangles perception and reasoning in Vision-Language Models (VLMs) for improved model assessment and efficient VLM development.

This paper introduces a Bayesian approach to setting the privacy budget in differential privacy, enabling agencies to balance data utility and confidentiality by customizing risk profiles.

Principled Probabilistic Imaging uses diffusion models as plug-and-play priors for accurate posterior sampling in inverse problems, surpassing existing methods.

Generative AI’s unique characteristics necessitate new pricing strategies; this paper models a sequential pricing game between competing firms, revealing the first-mover’s performance needs to be sign…

Noise Regularization rescues Deep Canonical Correlation Analysis from model collapse!

Orthogonal regularization prevents dimensional collapse in self-supervised learning, significantly boosting model performance across diverse benchmarks.

Random noise pretraining dramatically speeds up and enhances neural network learning without weight transport, mimicking the brain’s developmental process and achieving performance comparable to backp…

CoDA-NO, a novel neural operator, revolutionizes multiphysics PDE solving via codomain tokenization, enabling efficient self-supervised pretraining and few-shot learning for superior generalization.

Pretrained transformers surprisingly learn low-dimensional nonlinear functions efficiently from few in-context examples, outperforming baseline algorithms.

Pretrained Optimization Model (POM) excels at zero-shot black-box optimization, outperforming existing methods, especially in high dimensions, through direct application or few-shot fine-tuning.

PrefPaint: Aligning image inpainting diffusion models with human preferences using reinforcement learning, resulting in significantly improved visual appeal.

Eliciting high-dimensional probability distributions from experts using only preference comparisons is achieved via normalizing flows and a novel functional prior, resolving the problem of collapsing …

PreTS algorithm efficiently identifies the most preferred policy in bandit problems with vector-valued rewards, achieving asymptotically optimal sample complexity.

Human-like choice modeling revolutionizes preference learning! A new tractable model, CRCS, significantly improves utility inference from human data, outperforming existing methods.

Despite common belief, state-of-the-art preference learning algorithms for LLMs achieve surprisingly low ranking accuracy, highlighting significant flaws in current alignment techniques.

Preference Flow Matching (PFM) streamlines preference integration into pre-trained models using flow matching, overcoming fine-tuning limitations and enabling robust alignment with human preferences.

PCformer boosts Transformer performance by using a predictor-corrector learning framework and exponential moving average coefficient learning for high-order prediction, achieving state-of-the-art resu…