About
Research & Development
FastVLM: Accelerating Vision-Language Model Inference with Self-Speculative Decoding
TLDR: FastVLM is a new framework that speeds up Vision-Language Model (VLM) inference by 1.55-1.85x. It uses a Self-Speculative Decoding (SSD) approach with a...
Research & Development
E2RANK: A Unified Approach for Efficient and Effective Text Retrieval and Reranking
TLDR: E2RANK is a novel framework that transforms a single text embedding model into both an efficient retriever and a high-quality listwise reranker. It...
Research & Development
Predicting Commercial Memorability: How LLMs Fuse Multi-modal Data for Better Ad Recall
TLDR: A research paper introduces an LLM-based multimodal fusion system, called Gemma Fusion, for predicting commercial (brand) memorability. It integrates visual (ViT) and textual...
Research & Development
Controlling AI Image Generation with Semantic Surgery
TLDR: Semantic Surgery is a novel, training-free method for precisely removing unwanted concepts (like objects, explicit content, or artistic styles) from images generated by...
Research & Development
Decoupling Exploration and Safety for Robust AI Learning
TLDR: RLPD-GX, or GUARDIAN, is a novel framework for hybrid offline-online reinforcement learning that addresses instability by separating a reward-seeking 'Learner' from a safety-enforcing...
Research & Development
Adaptive Prediction for Self-Evolving AI: Quantifying LLM Confidence
TLDR: A new framework, AR-NECP, enhances the reliability of self-evolving large language models (LLMs) by addressing challenges in continual learning. It uses adaptive rejection...