Articles for category: AI Research

March 5, 2025

ikayaniaamirshahzad@gmail.com

Machine learning on interictal intracranial EEG predicts surgical outcome in drug resistant epilepsy

Ethics statement The protocol was approved by North Texas Regional IRB (2019-166; PI: C. Papadelis) that waived the need for informed consent considering the study’s retrospective nature. All methods and analyses were performed in accordance with relevant guidelines and regulations. Study cohort We retrospectively analyzed iEEG data recorded from 43 children and young adults with DRE who had resective neurosurgery at Boston Children’s Hospital between June 2011 and June 2018. We selected patients based on the following criteria: (i) availability of at least 5-minute interictal iEEG data; (ii) availability of post-implantation computerized tomography (CT); (iii) availability of preoperative and postoperative

March 5, 2025

ikayaniaamirshahzad@gmail.com

Partially rewriting an LLM in natural language

Our most recent work on using sparse autoencoders (SAEs) focused on automatically generating natural language interpretations for their latents and evaluating how good they are. If all the latents were interpretable, we could use the interpretations to simulate the latent activations, replacing the SAE encoder with an LLM and a natural language prompt. We should then be able to patch the activations generated by this natural language simulation back into the model and get nearly identical behavior to the original. In the limit, we could effectively “rewrite” the entire model in terms of interpretable features and interpretable operations on those

March 5, 2025

ikayaniaamirshahzad@gmail.com

Stanford CRFM

The Holistic Evaluation of Language Models (HELM) framework is an open source framework for reproducible and transparent benchmarking of language models that is widely adopted by academia and industry. To meet HELM users’ needs for more powerful benchmarking features, we are proud to announce our collaboration with Unitxt, an open-source community platform developed by IBM Research for data preprocessing and benchmark customization. The integration of Unitxt into HELM gives HELM users access to the vast Unitxt catalog of benchmarks, and allows users to run sharable and customizable evaluation pipelines with greater ease. Installation and Usage First, install HELM with the

March 5, 2025

ikayaniaamirshahzad@gmail.com

Language Models Reinforce Dialect Discrimination – The Berkeley Artificial Intelligence Research Blog

Sample language model responses to different varieties of English and native speaker reactions. ChatGPT does amazingly well at communicating with people in English. But whose English? Only 15% of ChatGPT users are from the US, where Standard American English is the default. But the model is also commonly used in countries and communities where people speak other varieties of English. Over 1 billion people around the world speak varieties such as Indian English, Nigerian English, Irish English, and African-American English. Speakers of these non-“standard” varieties often face discrimination in the real world. They’ve been told that the way they speak

March 5, 2025

ikayaniaamirshahzad@gmail.com

2.0 Flash, Flash-Lite, Pro Experimental

In December, we kicked off the agentic era by releasing an experimental version of Gemini 2.0 Flash — our highly efficient workhorse model for developers with low latency and enhanced performance. Earlier this year, we updated 2.0 Flash Thinking Experimental in Google AI Studio, which improved its performance by combining Flash’s speed with the ability to reason through more complex problems. And last week, we made an updated 2.0 Flash available to all users of the Gemini app on desktop and mobile, helping everyone discover new ways to create, interact and collaborate with Gemini. Today, we’re making the updated Gemini

March 5, 2025

ikayaniaamirshahzad@gmail.com

[2410.13770] Probing the Latent Hierarchical Structure of Data via Diffusion Models

[Submitted on 17 Oct 2024 (v1), last revised 28 Feb 2025 (this version, v2)] View a PDF of the paper titled Probing the Latent Hierarchical Structure of Data via Diffusion Models, by Antonio Sclocchi and 3 other authors View PDF HTML (experimental) Abstract:High-dimensional data must be highly structured to be learnable. Although the compositional and hierarchical nature of data is often put forward to explain learnability, quantitative measurements establishing these properties are scarce. Likewise, accessing the latent variables underlying such a data structure remains a challenge. In this work, we show that forward-backward experiments in diffusion-based models, where data is

March 5, 2025

ikayaniaamirshahzad@gmail.com

Enhancing Costmap for Off-Road Navigation with Robust IR-RGB Fusion for Improved Day and Night Traversability

[Submitted on 4 Dec 2024 (v1), last revised 2 Mar 2025 (this version, v2)] View a PDF of the paper titled IRisPath: Enhancing Costmap for Off-Road Navigation with Robust IR-RGB Fusion for Improved Day and Night Traversability, by Saksham Sharma and 2 other authors View PDF HTML (experimental) Abstract:Autonomous off-road navigation is required for applications in agriculture, construction, search and rescue and defence. Traditional on-road autonomous methods struggle with dynamic terrains, leading to poor vehicle control in off-road conditions. Recent deep-learning models have used perception sensors along with kinesthetic feedback for navigation on such terrains. However, this approach has out-of-domain

March 5, 2025

ikayaniaamirshahzad@gmail.com

Weak-to-Strong Alignment via Multi-Agent Contrastive Preference Optimization

[Submitted on 10 Oct 2024 (v1), last revised 2 Mar 2025 (this version, v2)] View a PDF of the paper titled MACPO: Weak-to-Strong Alignment via Multi-Agent Contrastive Preference Optimization, by Yougang Lyu and 6 other authors View PDF HTML (experimental) Abstract:As large language models (LLMs) are rapidly advancing and achieving near-human capabilities on specific tasks, aligning them with human values is becoming more urgent. In scenarios where LLMs outperform humans, we face a weak-to-strong alignment problem where we need to effectively align strong student LLMs through weak supervision generated by weak teachers. Existing alignment methods mainly focus on strong-to-weak alignment

March 4, 2025

ikayaniaamirshahzad@gmail.com

[2412.17762] The Superposition of Diffusion Models Using the Itô Density Estimator

[Submitted on 23 Dec 2024 (v1), last revised 28 Feb 2025 (this version, v2)] View a PDF of the paper titled The Superposition of Diffusion Models Using the It\^o Density Estimator, by Marta Skreta and Lazar Atanackovic and Avishek Joey Bose and Alexander Tong and Kirill Neklyudov View PDF HTML (experimental) Abstract:The Cambrian explosion of easily accessible pre-trained diffusion models suggests a demand for methods that combine multiple different pre-trained diffusion models without incurring the significant computational burden of re-training a larger combined model. In this paper, we cast the problem of combining multiple pre-trained diffusion models at the generation

March 4, 2025

ikayaniaamirshahzad@gmail.com

How Google Research is making healthcare more accessible and personalized with AI

Last week, at the Lake Nona Impact Forum for advancing global health, I discussed the potential of AI to meaningfully improve healthcare and advance science. Our recent AI breakthroughs provide unprecedented opportunities to make healthcare more accessible, personalized and effective for everyone, and to significantly accelerate scientific discovery. Here’s an update on our progress, how we’re collaborating with partners to bring AI to global healthcare settings and our recently announced AI co-scientist. AI is making accurate health information more accessible Google is often the first place people turn to when they are looking for answers to health-related questions, so we