Valence Labs’ Post

Super excited to talk and discuss deep learning for science and engineering at AMLD!

On my journey towards my PhD, I had the chance to participate at the OHBM 2022 Hackathon - a report has been published (including exploding brains!) The following projects were reported I have been part of: 1. I am grateful for my internship at Psychoinformatics Lab lead by Michael Hanke, learning about the Datalad software which allows to track data manipulation. At the Hackathon, the lab carried out a testing of a website representation of datasets called Datalad Catalog which allows for fast and detailed viewing of the data, while using little computer memory. 2. I also joined Omer Faruk Gulban on his project to explode fMRI brains - I went through his code in Julia and learned about it. 3. Finally, I also tested out Neuroscout, lead by Alejandro de la Vega, a no code website to analyse publicly available fMRI data, notably naturalistic experiments (like watching films while in brain scanners). Article: https://lnkd.in/edfhEA3r

Glad to be joining as a mentor at the GARN-CLS BIOHACK Hackathon. I'll be joining the hackathon with a very interesting challenge. Humans are naturally multi-modal learners, demonstrating the capacity to process and fuse information from various sources concurrently. In contrast, machine models are often constrained by their modality-specific nature, limiting their capacity to fully leverage complementary data sources. Our aim is to explore the fusion of genomics with CT scans and the fusion of genomics with MRI scans. We aim to also incorporate explainable AI techniques, ranging from attention-based to attribution-based methods. These techniques aim to enhance transparency and understanding in healthcare decisions by identifying the most influential features

Are you interested in using #MachineLearning in your applications? Check out this blog to learn how students from Stellenbosch University won first place at MATHack Western Cape 2022 by creating an app in MATLAB that uses AI to forecast sales #Hackathon #MLH

1st place at the Columbia University Hackathon. Realizing the gravity of our challenge, I decided to pull an all-nighter as the clock ticked down, making sure every moment counted. By 9:58 AM, just minutes before the submission deadline, we were in a frenzy, debugging and finalizing our project. 10:00 AM, we took a breath. Our achievement? YourCongress, a web app designed to simplify policies and help users make policy-driven choices instead of politician-driven ones. It simplifies finding, understanding, and influencing policies through an advanced search, powered by ChromaDB and OpenAI ADA embeddings, visualizing over 9500 interconnected policies. To ensure informed decisions, we integrated Bing AI Search for comprehensive policy data. Moreover, GenAI (OpenAI API) and Retrieval-Augmented Generation (RAG) allow users to craft and forward well-researched policy proposals directly to Congress and politicians. Winning felt great, but creating a killer product? That was the real victory. Shout out to my team, Devan Shah, Mahdi Tanzim, Fahad Faruqi

It's Day 2 of ARM Open Science Summer School! The 23 students are attending sessions led by ARM instructors at Cleveland State University to learn about ARM data and open science. If you're interested in attending #ARMSummerSchool in the future, stay up to date on all ARM events at ARM.gov! #ARMData #DOEClimateScience

What a wonderful conference has been this week in Edmonton! COLT2024 brought together some truly passionate researchers in machine learning theory, it was a true honour to be there. This picture shows me introducing the fascinating De La Valée Poussin kernel. This function is the foundation for some theorems I proved in my paper on reinforcement learning in continuous action spaces. Check it out here: "Projection by Convolution: Optimal Sample Complexity for Reinforcement Learning in Continuous-Space MDPs" by myself, Alberto Maria Metelli Matteo Papini Marcello Restelli link: https://lnkd.in/dC_YB3gY

A new format at the #HLF23 were the "Spark Sessions" – these are blocks that include several short plenary "Spark Talks" (15-20 minutes) by laureates. Here they give insights into their current research projects, discuss their work and spark inspiration for new ideas and problem-solving approaches. The session was moderated by Tom Crawford. Check out the first "Spark Session": # Raj Reddy: Speech to Speech Machine Translation # Yael Tauman Kalai: The Evolution of Proofs in Computer Science # Alexei Efros: Computing After the Victory of Data

Finally, the evaluation phase of the ADIA Lab competition is done, and I secured the 18th rank out of 4000 participants worldwide, which feels amazing 😎 ! Here are the key lessons I've learned from this experience: 1. Avoid Overfitting: Overfitting is a real challenge. While training, I was ranked 150th, but I've moved up to 18th by avoiding overfitting. It's crucial to follow proper procedures and assess your machine learning model accurately to prevent overfitting on the training data. 2. Focus on Feature Engineering: Creating the right features is vital. Proper feature engineering directly impacts your model's performance. 3. Ensure Model Stability: It's not just about high accuracy; a model needs to be stable to gain trust. Even with great accuracy, an unstable model won't be reliable. Ensure your model is stable and trustworthy. 4. Participate in Competitions: Competitions are incredible learning experiences. Always take part because they teach valuable lessons. A big thanks to CrunchDAO and ADIA Lab for organizing this insightful competition! #datascience #machinelearning