Vignesh Narayanan

Physics-Informed Neural ODEs with Scale-Aware Residuals for Learning Stiff Biophysical Dynamics Kamalpreet Singh Kainth, Prathamesh Dinesh Joshi, Raj Abhijit Dandekar, Rajat Dandekar, Sreedat Panat https://lnkd.in/dtRD_ide This paper tackles a critical challenge in using Neural ODEs for modeling stiff biophysical systems, specifically the difficulty in achieving stable and accurate long-term predictions. The core idea is to improve training stability and accuracy by incorporating physics-informed regularization with scale-aware residual normalization within a Neural ODE framework. The authors propose "PI-NODE-SR," which combines two key elements: 1. Physics-Informed Regularization: They leverage the known governing equations (e.g., Hodgkin-Huxley) to define a physics-informed loss term that penalizes deviations from the expected behavior. This is a standard PINN approach, but the key is how they handle the residual. 2. Scale-Aware Residual Normalization: This is the novel contribution. Stiff systems often involve state variables evolving on vastly different timescales. Directly applying a physics-informed loss can lead to one timescale dominating the training process, hindering convergence and accuracy. To address this, they normalize the residual associated with each state variable by a scaling factor. This factor is chosen to balance the contributions of each variable to the overall loss, preventing the faster dynamics from overwhelming the slower ones. They use a low-order explicit solver (Heun method) to calculate the residual. The authors demonstrate the effectiveness of PI-NODE-SR on the Hodgkin-Huxley equations, showing that it can learn from a single oscillation and extrapolate accurately over longer time horizons. Importantly, they show that the method can recover morphological features in the gating variables that are typically only captured by higher-order solvers. ----- This work directly addresses a significant limitation of Neural ODEs in scientific applications: their struggle with stiff systems. By introducing scale-aware residual normalization, the authors provide a principled way to stabilize training and improve the accuracy of long-term predictions. This is particularly relevant for: - Biophysical modeling: Accurately simulating neuronal dynamics, cardiac electrophysiology, and other complex biological processes. - Chemical kinetics: Modeling reaction networks with widely varying reaction rates. - Multi-physics simulations: Situations where different physical processes evolve on different timescales. The ability to use lower-order solvers with neural correction to achieve accuracy comparable to higher-order methods has significant implications for computational efficiency. While the method's sensitivity to initialization is noted, the overall approach offers a promising direction for developing more robust and efficient Physics-Informed Neural ODEs for a wide range of scientific applications.

146

1 Comment

Running LLMs on-device is quickly moving from theory to practice. This blog by Vikas Chandra is a great read on the state of on-device LLMs and why efficient architectures, model optimization, and local inference matter more than ever. https://lnkd.in/dzPRtnJt It’s also exactly what Hailo-10H was built for - enabling LLMs and VLMs to run directly on devices, with low latency, strong performance, and full data privacy. From Raspberry Pi AI HAT+ 2 to the ASUS UGen300 USB Edge AI accelerator and more, we’re seeing real generative AI running at the edge today.

27

Exciting news for India’s newsrooms! Google has launched the first-of-its-kind AI Skills Academy to help journalists and media organizations master AI tools. This program is all about responsible adoption, boosting efficiency, and unlocking new potential in storytelling. It’s inspiring to see this continuous investment in AI upskilling to empower talent across the industry. Check out the details here: https://lnkd.in/gMdFGjQd #ProudToBeGoogle #AIinJournalism #FutureofNews #Innovation Aiyappa M M Arpan Tiwari Shreekar Kavathalkar Hari Hara Eshwaran SNeelansh Verma Shankar Selvaraj Pooja Jha Kajal Ruia

19

1 Comment

NVIDIA's deep partnerships with research universities drive mutual innovation with key pillars: cutting-edge AI infrastructure, software, and funding through hackathons, grants, and joint research projects. Investing in the future of education and innovation creates a talent pipeline that benefits the entire tech ecosystem. Read more on @Yahoo: https://yhoo.it/428Xkpy @University-of-Toronto, @UofFlorida, @Georgia-Institute-of-Technology, @University-at-Albany

5

1 Comment

India launched a program to build native large language models for its many languages, funding startups and pooling compute under the $1.2 billion IndiaAI Mission. The Ministry of Electronics and Information Technology reserved 19,000 GPUs (including 13,000 Nvidia H100s). So far, it's backed Sarvam AI’s 70 billion-parameter multilingual model with voice and reasoning, plus projects from Soket AI Labs, Gan AI and Gnani AI. In The Batch, we break down the projects, model sizes, and implications for voice-first apps. Read the full article: https://hubs.la/Q03D21xv0

1,035

20 Comments

📢 We are excited to share a second collaborative blog between Clario and Amazon Web Services (AWS). Co-authored by Clario’s Kim Nguyen and Shyam Banuprakash with AWS’s Praveen Rao Haranahalli S, the article showcases how Clario is leveraging generative AI on Amazon’s cloud service platform, Bedrock, to streamline and scale clinical trial software configurations.     🧩 Building on our earlier post about Clario’s investment in AI to transform clinical trial documentation, this new article highlights how we're advancing study setup and configuration through AI-driven automation that improves efficiency, accuracy, and scalability across trials.      Read the full blog 🔗 https://lnkd.in/es7izf2y     Catch up on our earlier work here 🔗 https://lnkd.in/gPT39rTq     #ClinicalTrials #ClarioAI #AmazonBedrock #ResponsibleAI #GenerativeAI  

138

🔥 Day 11 of #30DaysofCode Today’s challenge: “Repeated DNA Sequences” 🧬 The problem asked me to find all 10-letter-long DNA subsequences that appear more than once in a string. 💡 Key learnings: A direct unordered_map<string,int> solution works but is slower due to string slicing + hashing. A bitmask approach (2 bits per nucleotide) allows encoding each 10-letter substring in just 20 bits. By using a rolling hash + a fixed-size array of size 1<<20, lookups are O(1) and extremely cache-friendly. Runtime dropped from ~0.64 ms (hash set approach) to ~0.02 ms with bitmask optimization ⚡. ✅ Takeaway: The right encoding and data structure can turn a good solution into a blazing-fast one. Consistency continues — 11 days strong 🚀 #30DaysofCode #LearningInPublic #Algorithms #Optimization #ProblemSolving

19

🧠💬 Memory in LLMs A practical guide showing how to implement conversational memory in LLMs using LangGraph, demonstrated through a therapy chatbot. Features code examples for basic retention, trimming, and summarization approaches. Learn to build memory-aware apps 👉 https://lnkd.in/gybcrV5v

968

21 Comments

Scaling AI is hard. Anyscale and Google Cloud make it easier. Read how Anyscale – built by the creators of Ray runs on Google Compute Engine to help teams scale any AI workload, from LLMs to classic ML. 👉 https://lnkd.in/gVr39dwZ

66

2 Comments

Empowering India’s newsrooms with AI! 💡Google has launched the AI Skills Academy, a first-of-its-kind initiative to help journalists and news organizations across India harness the power of AI responsibly and effectively. It’s inspiring to see how Google continues to invest in AI upskilling and innovation, empowering talent and transforming industries. Read more: https://lnkd.in/gGTfnqfn #ProudToBeGoogle #AI #Innovation #Journalism #DigitalSkills #GoogleNewsInitiative Sonia Aswani Vijay Nayak Ashish Sharma Tanzeem Ulla Nikhil Joseph Simmi Sarna Arpan Tiwari Neelansh Verma Pooja Jha Vipin V Vivek K S Nikhita Singh

23