3

MoGround: Measuring and Mitigating Modality Distraction in Vision-Language Models
We release MoGround, a vision-language dataset spanning four visual domains in which the answer to every question is guaranteed to be …
Cite PDF Code Dataset
MoGround: Measuring and Mitigating Modality Distraction in Vision-Language Models
The Undetected Damage of Quantization on Retrieval and How to Fix It
We show that a quantized model that keeps its classification accuracy still changes 14 to 46% of its top-1 retrieval results, and that …
Cite PDF Code
The Undetected Damage of Quantization on Retrieval and How to Fix It
HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation
Fully homomorphic encryption (FHE) allows a server to run a language model directly on encrypted user prompts, but current approaches …
Cite arXiv Code Model
HEAT: Faster Fully Homomorphic Inference via Approximations-Weights Co-Adaptation
Steering Vectors are an Adversarial Attack Surface
Activation steering has become a popular way to control Large Language Model (LLM) behavior without fine-tuning. Since the technique is …
Cite arXiv
Steering Vectors are an Adversarial Attack Surface
How Neural Losses Shape VAE Latents
Modern VAEs are rarely trained with the pointwise likelihood implied by the standard β-VAE objective. In practice, pointwise …
Cite arXiv PDF
How Neural Losses Shape VAE Latents
Communicating Sound Through Natural Language
Natural language is widely used to describe, prompt, and control audio systems, but rarely serves as the representation carrying audio …
Cite arXiv Demo
Communicating Sound Through Natural Language
Multi-objective Evolutionary Merging Enables Efficient Reasoning Models
Reasoning models have demonstrated remarkable capabilities in solving complex problems by leveraging long chains of thought. However, …
Cite arXiv
Multi-objective Evolutionary Merging Enables Efficient Reasoning Models
Domain Elastic Transform: Bayesian Function Registration for High-Dimensional Scientific Data
Nonrigid registration is conventionally divided into point set registration, which aligns sparse geometries, and image registration, …
Cite arXiv GitHub
Domain Elastic Transform: Bayesian Function Registration for High-Dimensional Scientific Data
Activation Patching for Interpretable Steering in Music Generation
Understanding how large audio models represent music, and using that understanding to steer generation, is both challenging and …
Cite arXiv
Activation Patching for Interpretable Steering in Music Generation
Decoding RNA-RNA Interactions: The Role of Low-Complexity Repeats and a Deep Learning Framework for Sequence-Based Prediction
RNA-RNA interactions (RRIs) are fundamental to gene regulation and RNA processing, yet their molecular determinants remain unclear. In …
Cite bioRxiv GitHub
Decoding RNA-RNA Interactions: The Role of Low-Complexity Repeats and a Deep Learning Framework for Sequence-Based Prediction