Feng Tan

🚀 Excited to share Compressive Light-Field Tokens (CLiFT) — a new scene representation for neural rendering developed by my students (Zhengqing Wang, Yuefan Wu, Jiacheng Chen, Fuyang Zhang). CLiFT is a Light-Field Token, compressed for compute-efficient and adaptive neural rendering, achieving (2x to 16x) reduction in storage and (66% to 100%) increase in FPS, while preserving high visual quality with only (0.54dB to 3dB) drop in PSNR. 🔗 Project: https://lnkd.in/gBqvMWi8 📦 Code coming 8/1!

166

3 Comments

TRI's "LBM 1.0" paper appeared on arxiv last night! Large Behavior Models (LBMs) are foundation models for robots that map robot sensors (notably camera inputs) and natural language commands into robot actions. The robots are programmed just through demonstration; we can develop incredible new skills like the video below without writing a single line of new code. There is a lot of excitement in the field right now because of the incredible potential for this type of technology. Inevitably, there is also a lot of hype. One of our main goals for this paper was to put out a very careful and thorough study on the topic to help people understand the state of the technology, and to share a lot of details for how we're achieving it. The short version is: LBMs work! We see consistent and statistically significant improvements as we increase the amount of pretraining data. But doing the science is still hard; as a field we have more work to do to improve the statistical power of our experiments. Please check out our project website for the paper and more details: https://lnkd.in/eDn_sqGh. https://lnkd.in/epSksw5E

769

9 Comments

🚀 Reflections on Six Years of PyBaMM – My Opening Plenary at the PyBaMM Conference Yesterday, I had the honor of delivering the opening plenary talk at the PyBaMM Conference, reflecting on six years of open-source battery modeling. PyBaMM started as an idea in 2018—"Physics-based models are accurate, but slow and hard to implement"—and has since grown into a widely used framework with 40K+ monthly downloads, 120+ contributors, and $20M+ in simulated experiments per month. Lessons from the Journey 🔹 Write tests & docs from day 1 – This provides a solid foundation on which to iterate quickly. 🔹 Release early and often – Community feedback accelerates progress. 🔹 The scientific method applies to software too – Iteration and validation are key. 🔹 Not knowing "how" can be an advantage – Innovation thrives on open problems. 🔹 You can’t please everyone – It's better for a few people to love it than many people to kind of like it Challenges in Industry Adoption 📊 Parameterization & customization – Companies need PyBaMM to fit their specific cells. 🐍 Python expertise – Many engineers are new to symbolic modeling. 🖥️ User interfaces & integration – Making PyBaMM more accessible remains a priority. With a growing community and increasing industry adoption, there’s still a lot of work ahead—BMS integration, next-gen chemistries, and more intuitive workflows. That’s where Ionworks comes in, helping to bridge PyBaMM with industrial applications. 🙌 Huge thanks to the PyBaMM team, contributors, and users who have made this journey possible! If you're using PyBaMM or interested in collaborating, let’s connect.

219

7 Comments

The Spring issue of IEEE Solid-State Circuits Magazine is out! https://lnkd.in/gxbjNtKw This issue is dedicated to Integrated Circuits for AI and ML applications, and was guest edited by Marian Verhelst and Charlotte Frenkel. It features four excellent articles that highlight core challenges and methodologies in AI hardware acceleration (see https://lnkd.in/gayteYmV for more detail). The feature articles are: 1) "Design and Optimization of Efficient Digital Machine Learning Accelerators" by Wei Tang, Sung-Gun Cho, Jie-Fang Zhang, and Zhengya Zhang examines AI acceleration through optimized computational kernels, such as matrix–matrix multiplication, leveraging architectures like SIMD and systolic arrays (10.1109/MSSC.2025.3549361). 2) "SRAM- and eDRAM-Based Compute-in-Memory Designs, Accelerators, and Evaluation Frameworks," by Yifan He, Xiaofeng Hu, Hongyang Jia, and Jae-sun Seo addresses in-memory computing (IMC) as a means to reduce data movement overhead, covering both analog and digital IMC approaches (https://lnkd.in/gHiukWKx). 3) "Open Source Heterogeneous SoCs for Artificial Intelligence," by Francesco Conti, Angelo Garofalo, Davide Rossi, Giuseppe Tagliavini, and Luca Benini explores the integration of heterogeneous systems on chip, demonstrating how open source platforms such as PULP support scalable AI hardware development (https://lnkd.in/gAQgK-8d). 4) " Sparsity-Aware Hardware," by Man Shi, Adrian Kneip, Nicolas Chauvaux, Jiacong Sun, Charlotte Frenkel, and Marian Verhelst discusses sparsity-aware hardware design, illustrating how structured and unstructured sparsity can be leveraged for significant performance improvements using an adapted roofline model (https://lnkd.in/gv8eTyvd). The issue also features regular columns: 1. Behzad Razavi’s “The Analog Mind” column (https://lnkd.in/gxKyqZr7) continues the series on CMOS inverter applications 2. Chris Mangelsdorf’s “Shop Talk: What You Didn’t Learn in School” column picks up where the previous issue left off, continuing the discussion on Miller compensation in "Miller’s Wrong Half-Plane Zero" (https://lnkd.in/gngTrywV). This time, the focus is on the right half-plane zero and its potential to degrade stability if not properly addressed. 3. Farhana S.’s “Circuits From a Systems Perspective” traces the historical evolution of wireless charging in everyday devices (https://lnkd.in/gnVhqUyF). Lot's of interesting articles to read! Enjoy! #ieeesscc #analog #ieee #ai #ml #solidstatecircuits #hardwareacceleration #semiconductors #inmemorycomputing

258

3 Comments

Imagine building a start-up on Deep Research system… without training a single model, and having the code open-sourced. 🤔 🚀 Excited to announce our new work from NVIDIA Research: Universal Deep Research (UDR): https://lnkd.in/gwRuU3Bw Deep research agents are one of the most widely used forms of AI agents. With UDR, we take a step forward: 🔍 Customizable strategy 🔛 research is not just about retrieving documents, but about shaping how the agent explores, verifies, and synthesizes knowledge. UDR shows that the research strategy itself can be adapted to the user’s goals. 🛠️ System-level behavior 🔛 the agent’s “deep research” workflow can be implemented without re-training or fine-tuning any LLMs. Instead, it’s orchestrated at the system level, making it lightweight, flexible, and easy to extend. 🌐 General-purpose design 🔛 UDR is universal: it can be applied across domains, whether for scientific discovery, technical due diligence, or business intelligence. AI agents are not just about better models - they are about better systems. We believe UDR opens the door to a new wave of startups and applications: anyone can now design a specialized “research assistant” that adapts to their domain and strategy. Learn more from products we offer in this domain: 🔗 AIQ Deep Research blueprint: https://lnkd.in/g2_4y3RR 🔗 NeMo Agent Toolkit: https://lnkd.in/gVPHTrzN Great project led by Peter Belcak.

126

3 Comments

🚍 What if shifting a bus departure by just a few minutes could save an entire vehicle? That’s the idea behind our new paper, recently published in the INFORMS Journal on Computing. Together with Thomas van der Schaft, we developed a Dynamic Discretization Discovery (DDD) algorithm that tackles the Multi-Depot Vehicle Scheduling Problem with Trip Shifting (MDVSP-TS). Why this matters: - Even tiny timetable adjustments can unlock huge efficiency gains in public transport. - Our algorithm solves real-world bus networks with nearly 4,000 trips to optimality. - It outperforms existing exact methods by a wide margin. This work shows how smart optimization can make public transport more efficient, sustainable, and practical at scale. 👉 Read more here: https://lnkd.in/eS_dhZZ3

93

9 Comments

Thrilled to be co-organizing the Symposium on Robotics & Automation in Smart Manufacturing Systems with @Haseung Chung, Amit Bagchi, Guangze Li, and Gary Zhang at ASME MSEC 2026 (June 22–26, Penn State, USA)! This symposium will spotlight how robotics, AI, digital twins, and automation are transforming manufacturing. From cobots to mobile and reconfigurable platforms, robotics now extends beyond traditional automation to enable flexible task execution, real-time decision-making, and intelligent human-machine interaction. Our goal is to highlight cutting-edge innovations and applications, while offering a forum for researchers and practitioners to exchange ideas on adaptive, resilient, and smart manufacturing systems. 📄 Details in the attached Call for Papers. ✨ Join us in shaping the future of smart manufacturing systems!

61

1 Comment

IEEE Robotics and Automation Society journals now use a double-anonymous review process! I haven't seen much discussion of this, but it is a pretty big change. This idea has been studied carefully in other fields, and while there is mixed evidence on if it helps with gender or racial bias, there are two main benefits which are well supported: 1) It makes a big difference in reducing bias towards well known labs/institutions/countries, as well as bias against less well known labs/institutions/countries. 2) It improves peoples perception of the fairness of the process. Researchers who feel that the process is biased against them may be stressed by the process or leave academia entirely. To be honest, I think the resistance to changes like this is largely laziness. It does take a little bit of effort to rephrase "In our prior work [X]" to "In [X]", but it shouldn't matter who wrote that article. In robotics, we often have school logos on the robots or in the background of images that we need to cover up. But once people get used to these standards it is a very minor burden. Furthermore, as reviewers I think we are often lazy and rely on what we know about the authors to influence our review in a way that is not in the best interest of science. So thank you, IEEE RAS, for making this change and I hope the community embraces it. RAS journal rules: https://lnkd.in/ehYhJA4U My lab's letter summarizing this research from a few years ago: https://lnkd.in/eGv4bBtF #robotics #robot #doubleblind #doubleanonymous #academia #paperreviews #journal

426

13 Comments

The final paper from Rachel Lim's time at Penn State was just published https://lnkd.in/egHb59wy. This work grew from nearly a decade of collaboration with Thien Phan. At the time, we were interested in using X-ray diffraction as a thermomechanical diagnostic for additive manufacturing, but it wasn't clear how we would ever be able to separate the effects of thermal and mechanical loading in the measured signal. With the help of Zi-Kui Liu and Shun-Li Shang, we trained a probabilistic machine learning model to deconvolute these two effects within experimental diffraction data (measured with Andrew Chuang at the Advanced Photon Source). While machine learning may not solve every problem, (when trained properly) it can be a valuable tool for solving complex inverse problems.

65

3 Comments