Advancing AI-Driven Neuroscience Research Through NVIDIA Connect Program
IIT Madras Brain Centre and Generative AI
Leading healthcare organizations, including the IIT Madras Brain Centre, are leveraging generative AI to enhance research capabilities. This center utilizes AI to analyze human brains at a cellular level across various demographics. They have developed a knowledge exploration framework integrating visual question-answering (VQA) models and large language models (LLMs) to facilitate the accessibility of brain imaging data for researchers.
Image courtesy of NVIDIA Technical Blog
Knowledge Exploration Framework
The framework allows researchers to connect brain imaging data with the latest neuroscience publications. The process is divided into two parts:
- Ingestion: This part indexes the latest neuroscience publications into the knowledge base.
- Q&A: Enables user interaction through queries, facilitating the exploration of neuroscience advancements related to specific conditions.
The Q&A section incorporates a multimodal retrieval-augmented generation (RAG) pipeline, which enhances user engagement with both text and images, ensuring relevant content is prioritized.
Improving Research Challenges with NVIDIA Technology
NVIDIA’s technology stack supports the framework’s processing pipeline, improving retrieval accuracy significantly. Fine-tuning of embedding models through NVIDIA NeMo and the use of the Mixtral 8x 7B NVIDIA NIM microservice for fast LLM inference have collectively enhanced top-2 retrieval accuracy.
User Input Filtering
To maintain the relevance of user input, NVIDIA NeMo Guardrails are utilized. A custom prompt tailored for neuroscience was tested, achieving a 98% acceptance rate for neuroscience-specific questions. This ensures that irrelevant questions are blocked effectively.
NVIDIA AI Blueprint for PDF Data Extraction
The recently launched NVIDIA AI Blueprint for multimodal PDF data extraction provides a framework for accurately extracting relevant information from neuroscience publications in PDF formats. This model aims to improve the efficiency of knowledge extraction from complex documents.
Image courtesy of NVIDIA Technical Blog
ConductScience’s Partnership with NVIDIA
ConductScience has joined the NVIDIA Connect program, which will accelerate the development of solutions like ConductVision, ConductInsecta, ConductNest, and GaitMaster. This partnership allows ConductScience to access GPU-accelerated computing, enhancing research on locomotion and cognition.
Innovations in Behavioral Research
ConductScience provides advanced technologies for behavioral research, including:
- ConductVision: A tracking suite for rodent behavior analysis.
- ConductInsecta: A platform for rapid insect identification.
- ConductNest: A home-cage monitoring system for behavioral observation.
- ConductScience GaitMaster: A solution for gait analysis in animal models and humans.
These tools enhance research efficiency, enabling scientists to gather precise data on behavior.
Firefly Neuroscience and the NVIDIA Connect Program
Firefly Neuroscience, Inc. has also joined the NVIDIA Connect program, focusing on building a foundation model of the human brain utilizing their Brain Network Analytics (BNA) technology. This platform aims to advance diagnostic methods for mental health disorders.
Firefly’s BNA platform allows comprehensive brain health evaluations through advanced data analysis, combining EEG data with behavioral performance metrics.
