Posters

PuLID: Lightning-fast, tuning-free ID customization for text-to-image!

Leveraging public data enhances differentially private (DP) learning, but its limits are unclear. This paper establishes tight theoretical bounds for DP stochastic convex optimization, revealing when …

PTQ4DiT achieves 8-bit and even 4-bit weight precision for Diffusion Transformers, significantly improving efficiency for image generation without sacrificing quality.

Pairwise Softmax Loss (PSL) improves recommendation accuracy by enhancing Softmax Loss (SL) with alternative activation functions, resulting in tighter ranking metric surrogates and better noise resis…

SelfFormer: A novel self-supervised transformer-based method outperforms existing techniques by leveraging directional self-attention for efficient and accurate real-world image denoising.

Pseudo-Private Data Guided Model Inversion (PPDG-MI) significantly improves model inversion attacks by dynamically tuning the generative model to increase the sampling probability of actual private da…

DASH: a novel pruning framework leverages domain knowledge to improve the interpretability and sparsity of neural network models for gene regulatory dynamics, outperforming existing methods.

Prune and Repaint: A new content-aware method for superior image retargeting across any aspect ratio, preserving key features and avoiding artifacts.

ProxyFusion, a novel face feature fusion method, achieves real-time performance by using sparse experts to weight features without relying on intermediate representations or metadata, substantially im…

Unmeasured confounders hinder causal inference; this paper introduces a novel method using two pre-treatment text instances and zero-shot models to infer proxies for unobserved confounders, enabling p…

ProvNeRF enhances NeRF reconstruction by modeling per-point provenance as a stochastic field, improving novel view synthesis and uncertainty estimation, particularly in sparse, unconstrained view sett…

POETRY: a recursive neural theorem prover achieving 5.1% higher success rate and solving substantially longer proofs.

Transformers excel at in-context learning (ICL), solving new tasks with just prompts. This paper provides a mathematical explanation, showing how transformers use multi-concept word semantics to achie…

Verifiably safe AI controllers are created via a novel framework, VerSAILLE, which uses differential dynamic logic and open-loop NN verification to prove safety for unbounded time horizons.

Provably robust diffusion posterior sampling for plug-and-play image reconstruction is achieved via a novel algorithmic framework, DPnP, offering both asymptotic and non-asymptotic performance guarant…

Researchers achieve provably optimal memory capacity in transformer-compatible Hopfield models by framing the problem as an optimal spherical code arrangement, resulting in a novel sublinear time algo…

RLHF’s overoptimization problem is mitigated by RPO, a novel algorithm that uses SFT loss as an implicit adversarial regularizer, ensuring efficient and effective LLM alignment.

Faster bilevel optimization is achieved via without-replacement sampling, improving convergence rates compared to independent sampling methods.

This paper presents novel RL algorithms using multinomial logit function approximation, achieving O(1) computation and storage while nearly closing the regret gap with linear methods.

Provably efficient algorithms are introduced for interaction-grounded learning (IGL) with context-dependent feedback, addressing the lack of theoretical guarantees in existing approaches for personali…