Search

Results for "SimulMask"

Simultaneous Masking, Not Prompting Optimization: A Paradigm Shift in Fine-tuning LLMs for Simultaneous Translation

arXiv · May 16

This paper introduces SimulMask, a new paradigm for fine-tuning large language models (LLMs) for simultaneous translation. SimulMask utilizes a novel attention masking approach that models simultaneous translation during fine-tuning by masking attention for a desired decision policy. Applied to a Falcon LLM on the IWSLT 2017 dataset, SimulMask achieves improved translation quality compared to state-of-the-art prompting optimization strategies across five language pairs while reducing computational cost. Why it matters: The proposed method offers a more efficient way to adapt LLMs for real-time translation, potentially enhancing multilingual communication tools and services.

An AI trained to spot hidden objects can see through camouflage - New Scientist

Inception · May 12

Researchers at the University of Maryland have developed an AI system that can identify objects hidden by camouflage. The AI uses a convolutional neural network trained on synthetic data to detect partially occluded objects. The system outperformed existing object detection methods in tests on real-world images. Why it matters: The work demonstrates potential applications of AI in defense, security, and search and rescue operations in the Middle East and elsewhere.

A Culturally-diverse Multilingual Multimodal Video Benchmark & Model

arXiv · Jun 8

A new benchmark, ViMUL-Bench, is introduced to evaluate video LLMs across 14 languages, including Arabic, with a focus on cultural inclusivity. The benchmark includes 8k manually verified samples across 15 categories and varying video durations. A multilingual video LLM, ViMUL, is also presented, along with a training set of 1.2 million samples, with both to be publicly released.