​Posted by Arsha Nagrani and Paul Hongsuck Seo, Research Scientists, Google Research Automatic speech recognition (ASR) is a well-established technology that is widely adopted for various applications such as conference calls, streamed video transcription and voice commands. While the challenges for this technology are centered… Read More »AVFormer: Injecting vision into frozen speech models for zero-shot AV-ASR Google AI Google AI Blog

​ Datasets are often drawn from various domains while training language models (LMs). For instance, a sizable publicly accessible dataset called The Pile has 24% online data, 9% Wikipedia, 4% GitHub, etc. The makeup of the pretraining data significantly impacts how well an LM performs.… Read More »Stanford and Google Researchers Propose DoReMi: An AI Algorithm Reweighting Data Domains for Training Language Models Aneesh Tickoo Artificial Intelligence Category – MarkTechPost

​ LLaMaTab – An Insightful Chrome Extension A Chrome add-on called LLaMaTab New Tab will display a different image of a llama every time a new tab starts. It’s a silly add-on, but it can keep one going when things become tough. LLaMaTab New Tab… Read More »Meet LLaMaTab: An Open-Source Chrome Extension that Runs an LLM Entirely in the Browser Dhanshree Shripad Shenwai Artificial Intelligence Category – MarkTechPost

​ Researchers from MIT and IBM Research have developed a tool called saliency cards to assist users in selecting the most appropriate saliency method for their specific machine-learning tasks. Saliency methods are techniques used to explain the behavior of complex machine learning models, helping users… Read More »MIT Researchers Introduce Saliency Cards: An AI Framework to Characterize and Compare Saliency Methods Niharika Singh Artificial Intelligence Category – MarkTechPost

​ Defending against deepfakes involves work on both detection and generating approaches. People that need to fortify their pipeline with additional functionality without creating complex boilerplate code can take advantage of its loose coupling nature. However, existing deepfake algorithms need better performance and a clearer… Read More »Meet DeepFaceLab: A Real-time Face Swap for PC streaming or Video Calls Dhanshree Shripad Shenwai Artificial Intelligence Category – MarkTechPost

​ Large Language Models (LLMs), the deep learning-based highly efficient models, are the current trend in the Artificial Intelligence community. The well-known chatbot developed by OpenAI, ChatGPT, is based on GPT architecture and has millions of users utilizing its abilities for content generation. Its incredible… Read More »How to Keep Scaling Large Language Models when Data Runs Out? A New AI Research Trains 400 Models with up to 9B Parameters and 900B Tokens to Create an Extension of Chinchilla Scaling Laws for Repeated Data Tanya Malhotra Artificial Intelligence Category – MarkTechPost

​ Computer vision has become increasingly important in industrial applications, serving product line management, stock control, and safety monitoring functions. However, utilizing computer vision at the edge of a network poses challenges, particularly regarding latency and reliance on mixed networks or cloud resources. To address… Read More »Meet Microsoft’s Open-Sourced KubeAI Application Nucleus: A Solution Accelerator for Creating, Deploying, and Operating Environment-Aware Solutions at Scale that Use Artificial Intelligence (AI) at the Edge Niharika Singh Artificial Intelligence Category – MarkTechPost

​Multimodal learning is defined as learning over multiple heterogeneous input modalities such as video, audio, and text. In this work, we are concerned with understanding how models behave as the type of modalities differ between training and deployment, a situation that naturally arises in many… Read More »Robustness in Multimodal Learning under Train-Test Modality Mismatch Apple Machine Learning Research

​*= Equal Contributors Applications of large open-domain knowledge graphs (KGs) to real-world problems pose many unique challenges. In this paper, we present extensions to Saga our platform for continuous construction and serving of knowledge at scale. In particular, we describe a pipeline for training knowledge… Read More »Growing and Serving Large Open-domain Knowledge Graphs Apple Machine Learning Research

​Modern neural networks are growing not only in size and complexity but also in inference time. One of the most effective compression techniques — channel pruning — combats this trend by removing channels from convolutional weights to reduce resource consumption. However, removing channels is non-trivial… Read More »Unconstrained Channel Pruning Apple Machine Learning Research