Atharva Joshi’s Post

Learning new stuff is a must in this increasingly rapid technology ecosystem. But learning how to learn is tough. We don’t really do it right and tend to apply “neuromyths” that don’t quite help us. Even though my main research and work topic is AI, my responsibility at Adigital is around all types of emerging technologies. Quantum computing is one of them, and while I had already read a book and a few articles on the topic some years ago, I realized I did not remember some of the most relevant elements. A while back I found Quantum Country, by Andy Matuschak and Michael Nielsen, a web page that explains the concepts of quantum computing and mechanics, but using spaced repetitions embedded in the web page. After signing up, the system sends you prompts to review the concepts you’ve been reading during the last weeks, helping you remember them in a deeper way. Research-validated techniques such as spaced repetitions, varied conditions or interleaving practices help learn better and faster, and it’s great to see these approaches in real projects. https://quantum.country/

Isn’t GAN a interesting thing which has far reaching applications!! There are still few challenges such as training GANs in the quantum domain which can be computationally very expensive and requires sophisticated design of the loss function to suit quantum processes. Integrating GANs with quantum hardware to harness the full power of quantum computing for state reconstruction and simulation is also complex as of now but who knows! Generative Adversarial Networks using Quantum Simulation: An experiment https://lnkd.in/gdEz35r2 Awarded for exemplary work in "The Sustainability, Feasibility, and Responsibility of GANs" where the assessment includes the technical, social, and economic impact of AI business technologies and employed GANs to create a product for the good of society.

Inspired by a recent review paper by Prof. Voss, I just published a new post on how machine learning is revolutionizing computational chemistry, specifically through enhancing Density Functional Theory (DFT). Check out the full article on my Substack for insights into this cutting-edge integration. https://lnkd.in/dDmti5Fm #MachineLearning #ComputationalChemistry #ScienceInnovation #TechTrends #DataScience #MaterialScience #DFT

Researchers have unveiled a groundbreaking innovation: a hybrid computational system powered by the dynamic Belousov-Zhabotinsky (BZ) reaction. This cutting-edge technology intertwines physical chemistry with computation, creating a platform that emulates complex behaviors seen in natural systems. With a 3D-printed grid of interconnected reactors, this system transcends traditional computing paradigms, offering new pathways for solving complex optimization problems and enhancing deep learning tasks with its inherent non-linear characteristics. #Chemistry #ComputationalChemistry  

Please help my PhD candidate at ANU School of Computing in ANU College of Engineering, Computing & Cybernetics with a survey of those using #MachineLearning or #DeepLearning (ML/DL) frameworks like TensorFlow, PyTorch, CNTK, MNXT, KERAS etc. Please distribute this message and consider participating yourself! #TechDebt is a prevalent issue in (ML/DL) systems. A recently uncovered type is #AlgorithmDebt. Iko-Ojo Simon is researching this and would value insights from ML/DL practitioners. More information about the survey: https://lnkd.in/gd9MAbdN

Join us May 10th for invited talks and discussions at the FRP Reinforcement Learning Symposium at the Hariri Institute for Computing, Boston University. During the day-long event featuring speakers from academia and industry, we will uncover fundamental challenges in reinforcement learning frameworks and directions toward addressing them, particularly toward closing the current gap between theory, AI model training, and real-world applications and users. Hosted by Boston University College of Engineering Professors Eshed Ohn-Bar (ECE, CS) and Alex Olshevsky (ECE, SE, CS). Speakers: Bahman Gharesifard, Professor, Electrical & Computer Engineering, University of California, Los Angeles amin karbasi, Associate Professor, Electrical Engineering & Computer Science, Yale University & Staff Research Scientist at Google Alec Koppel, AI Research Lead/VP in the Multiagent Learning and Simulation Group within Artificial Intelligence Research, JPMorgan Chase & Co. Na Li, Winokur Family Professor, Electrical Engineering and Applied Mathematics, Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) Antonin Raffin, Research Engineer, Robotics and Machine Learning, German Aerospace Center (DLR) Alejandro Ribeiro, Professor, Electrical and Systems Engineering (ESE), University of Pennsylvania Daniel Russo, Associate Professor, Decisions, Risk, and Operations Division, Columbia Business School Kaiqing Zhang, Assistant Professor (ECE), Institute for Systems Research (ISR), ECE UMD Learn more & register here: https://lnkd.in/enSSKzZu #ReinforcementLearning #RL #AI #ArtificialIntelligence #Optimization

🚨 New Paper Alert! 🚨 ✨ Our paper entitled “Repeatability of Fine-Tuning Large Language Models Illustrated Using QLoRA,” has been published by IEEE Access! ✨ 📜 In this paper, we study the challenges of repeatability in fine-tuning four LLMs using two public datasets with QLoRA. To assess repeatability, we fine-tune each LLM seven times under identical hardware and software conditions. Our findings underscore the impact of non-repeatable fine-tuning of LLMs in obtaining reproducible LLMs models and in understanding the behaviour of LLMs. A huge thanks to the great team Prof. Larry (Lawrence) Hall, Prof. Dmitry Goldgof, and Prof. Peter R. Mouton, Ph.D. and to USF, Nvidia, NU and NSF for the computing resources and funding support! 🙏 Read more here: https://lnkd.in/dzpEx9qe #Research #AI #DeepLearning #LLMs #IEEE #QLoRA #Repeatability #Reproduciblity

Excited to share that my paper, based on the results and contributions of my Master's thesis, has been accepted in the 'Computing Frontiers 2024' conference proceedings! 🎉 'Performance evaluation of acceleration of convolutional layers on OpenEdgeCGRA' https://lnkd.in/duFYPHb3 In this work, we addressed the need to efficiently deploy deep learning solutions on the edge. Specifically, we focused on exploring the efficient mapping of convolutional layers on the OpenEdgeCGRA, a low-power Coarse-Grain Reconfigurable Array. Our study explored various mapping techniques in order to find the best one, emphasizing factors like memory usage, latency, and energy efficiency. I want to thank all of my supervisors Juan Sapriza, Pasquale Davide Schiavone, Daniele Jahier Pagliari, David Atienza, Maurizio Martina, Alessio Burrello for their support, which allowed me to embark on this incredible experience. Grateful for the opportunity to contribute to the advancement of edge computing! #DeepLearning #EdgeComputing #Research

Exciting news ✨ I recently participated in the online workshop organized by the student branch of the IEEE Computational Intelligence Society at Rajalakshmi Engineering College. Titled “Vision X: Beyond the Horizon,” the workshop provided a comprehensive overview of the latest trends in computer vision across various industries. It covered fundamental principles of deep learning in computer vision and emphasized the critical role of data in driving successful applications. Additionally, you were introduced to advanced techniques like infrared imaging and 3D reconstruction. Your enthusiasm to apply these newfound skills to future projects reflects your commitment to continuous learning and growth in this exciting field. 🌟🔍👓 #rec #computervision #3D

🧠 do you think computers can mimic human brains? I have recently been fascinated by the idea of #neuromorphic computing and SNN’s which are inspired from biological neural systems, especially in the way that they perform computation. ❗️Got introduced to this topic, thanks to Daniel Hulme who was discussing machine consciousness and the fascinating idea of super intelligence… If evolution of nueromorphic systems and its ecosystem happens at scale, it feels like we may closer to start mimicking human brains, and provide machines a more efficient (low energy, high performance) way to learn and synthesize information. Jon Hammant what do you think ?🤔 #aws