Li Deng

Congratulations Pia Saunders Campbell, Merve Ayyüce KIZRAK, Ph.D., and Elsie Jang on the publication of "Meeting the AI Workforce Challenge." This blend of public-private-philanthropic perspectives, is a great read for anyone engaged in workforce development, AI policy, or the future of talent. As with any of the prior technological breakthroughs, we must and will evolve into the new jobs and skills of the future; however, the rapid pace of AI calls on us to do this faster than ever before. Key Takeaways 👇 🔄 AI is rapidly reshaping the U.S. labor market, particularly in entry-level and technical roles. Adapting quickly is urgent for both organizations and individuals. 🤝 The federal government's AI Action Plan and America’s Talent Strategy are solid starts, but coordinated cross-sector action—including youth leadership and local partnerships—is essential for equitable, future-ready outcomes. 📚 There’s a pressing need to integrate AI literacy, ethical reasoning, and problem-solving into all levels of education and professional training, supporting resilient, adaptive workers. 🪴 Youth engagement and inclusion are critical. Policies shaped with (and by) young people will be more innovative and enduring as the pace of technological change accelerates. Grateful for the opportunity to share my perspectives with this team along the way and learn alongside such passionate change makers! https://lnkd.in/eKfNX7UC #AI #WorkforceDevelopment #FutureofWork #Collaboration #TalentStrategy

21

��� A key challenge to solving #InverseProblems, especially in science and engineering settings, is limited data and compute for simulations. 🥧 The #ICLR2025 work of Apivich Hemachandra, Gregory Lau, See-Kiong Ng, and Bryan Kian Hsiang Low, PIED, is an #ExperimentalDesign framework for #PDE inverse problems that efficiently optimizes for the best observation inputs. 📌 PIED utilizes #PINNs to perform forward simulation and to solve the inverse problems, which allows for higher efficiency over numerical simulators and allow for meshless and differentiable simulations. 🤖 Using PINNs along with differentiable observation selection criteria, PIED selects the optimal design parameters for one-shot deployment, while exploitation parallel computation and gradient-based optimization methods. 📈 These benefits allow PIED to outperform existing ED benchmarks in IPs for both finite-dimensional and function-values inverse parameters, as demonstrated in various settings including on real experimental data. Check out our #ICLR2025 Poster #28 on 26 Apr in Poster Session 6 Sat 3pm Hall 3 + Hall 2B. Paper: https://lnkd.in/gDxav2Dn GitHub: https://lnkd.in/gZXDkA5H

31

1 Comment

🚀 Exciting News! Our New Paper on helping Multimodal LLMs see Beyond Language Prior got accepted to CVPR Workshop on Visual Concepts! 🚀 Ever wonder why Multimodal LLMs (MLLMs), despite seeing images, sometimes miss the bigger picture or rely too much on text? While text-only LLMs get rich feedback from every token, MLLMs often face a "sparse feedback" problem. They struggle to learn about image concepts not explicitly mentioned in text descriptions and can default to just predicting text based on language patterns, rather than truly seeing. 🔥 Our latest research tackles this head-on! We've developed novel training strategies that: 1️⃣ Deepen Visual Understanding using Visual Loss: Teach the MLLM to build a much richer internal representation of ALL visual concepts in an image. 2️⃣ Boost Visual Attention using Blank Tokens: Encourage the model to pay significantly more attention to what it sees by subtly weakening its over-reliance on predicting just from previous text tokens. 💡 The Impact? We're seeing strong performance improvements on demanding visual tasks in both upstream (core understanding) and downstream (application-level) settings! 🏆 Key Achievement: Our approach enables our Llama 3.1 8B based Llava model – which is smaller, uses a simpler model architecture, and operates on lower-resolution visual inputs – to MATCH the performance of the much larger Llama 3.2 11B on challenging visual reasoning benchmarks! 🤯 This shows the power of smarter training recipe! This is a crucial first step towards MLLMs that are more visually intelligent, reliable, and truly understand the world around them. Please stay tuned for our upcoming updates on model architecture advancements more conducive to capturing visual and language concepts as well as post training MLLMs to reason through tricky questions. 👉 Dive into the details! Read the full paper here: https://lnkd.in/gZcbw5yC A huge thank you to my incredible co-authors Aarti Ghatkesar Uddeshya Upadhyay and everyone who supported this journey! #MultimodalLLM #MLLM #VisualGrounding #VisCon

69

2 Comments

I’m truly honored to be invited by the IEEE BioSensors 2026 organizing team to deliver a tutorial on: Practical AI/ML for Biosignals: Bridging Statistical Features and Multi-Task Learning Biosignals are rich, messy, and highly variable. In this tutorial, we offer a practical entry point into AI/ML for biosensing by starting from what makes signals like electroencephalography (EEG) and photoplethysmography (PPG) different from typical datasets: their frequency–magnitude structure, noise and artifacts, and strong variability across people and hardware (often experienced as domain shift). We show how this domain knowledge should shape the way you represent signals and choose models. We then use statistical analysis as the bridge from raw waveforms to meaningful features, connecting intuitive signal properties to time-domain statistics and frequency-domain summaries that often carry the most useful information. From there, we explain how classical machine learning uses these engineered features, how deep learning can learn representations directly from raw inputs, and how hybrid pipelines can combine both worlds. We also discuss practical, lightweight approaches to interpretability and explainability that help researchers and practitioners trust and debug biosignal models. Finally, we focus on multi-task learning as an extension for many biosensing problems. We cover how to define tasks that help rather than fight each other, why multi-loss training can become unstable, and how adaptive loss weighting can reduce manual tuning and keep training balanced. Participants will leave with concrete pitfalls to avoid and practical takeaways they can apply immediately.

23

💥 ICML-25 Paper Alert! Introducing the JETTS benchmark, Judge Evaluation for Test-Time Scaling ⏲️ LLM-judges specialize in evaluating responses generated by other LLMs. Test-time scaling requires multiple model responses to be evaluated for the best one. So, can LLM-judges help test-time scaling? Check this thread to find the answer and some interesting analysis👇 Repo: https://bit.ly/3YteE6Z X: https://lnkd.in/gXvQsd4C Paper: https://bit.ly/4jtRGVM

105

New work on Rethinking Thinking Tokens: LLMs as Improvement Operators: https://lnkd.in/errhNuuz Reasoning training encourages LLMs to produce long chains of thought (CoT), improving accuracy via self-checking but increasing context length, compute cost, and latency. This work studies whether frontier models can achieve better trade-offs, higher accuracy with lower cost. The paper develops a simple yet effective Parallel-Distill-Refine (PDR) procedure: Generate diverse drafts in parallel, Distill them into a compact textual workspace, and Refine conditioned on this workspace. This decouples context length from total token count, allowing control over compute via parallelism. PDR yields higher accuracy than long CoT at lower latency. Training an 8B model with RL to align with PDR further shifts the Pareto frontier. On math benchmarks, PDR achieves +11% (AIME 2024) and +9% (AIME 2025) over single-pass baselines. With Lovish Madaan, Aniket Didolkar, Suchin Gururangan, John Quan, Ruan Silva, Manzil Zaheer, Sanjeev Arora, and Anirudh Goyal.

78

1 Comment

An extremely interesting area of LLM research is improving model quality during inference, exemplified by the recent Test-Time Reinforcement Learning (TTRL) paper from Yuxin Zuo et al. The researchers show critical improvements across multiple benchmarks by using inference-time data to fine-tune the model. Specifically, they adapt the model's weights toward a consensus answer derived from multiple responses it generates itself. This research is fascinating to me because it resembles how we humans constantly train our own brains through interactions with the world, not just during dedicated learning times.

12

1 Comment

One of the most underappreciated pain points in LLM RL training is policy staleness. When your generation and training loops run asynchronously — which they must at scale — the policy that generated your rollouts drifts from the policy you're actually updating. Most teams paper over this with token-level clipping (PPO-style) or just accept the instability. VESPO from RedNote takes a different approach: sequence-level importance weight reshaping with a closed-form variational solution. No length normalization hacks. The key claim — stable training even at 64x staleness ratios — is striking, and they show it holds for both dense and MoE architectures. From my team's work on RL post-training, I can say that the staleness problem is very real and gets worse as you scale. We've seen training runs destabilize when the generation queue backs up and policy lag exceeds ~8-10x. The usual fix is aggressive clipping or just throwing away stale batches, both wasteful. A principled sequence-level solution that handles extreme staleness would genuinely change how we architect async RL pipelines. What I'd want to see next: how VESPO interacts with reward model staleness (not just policy staleness), and whether the variance reduction holds up on tasks with sparser rewards like long-horizon coding or multi-turn dialogue. Paper: https://lnkd.in/g-X2DvGb #LLM #ReinforcementLearning #RLHF #MachineLearning #AIResearch #PostTraining

10

Useful read: Circuit tracing is crucial to understanding how LLMs are behaving behind the scenes. An article from MIT Tech Review covers some of the developments around circuit training. The snippet below explains how Claude is performing basic math. (https://lnkd.in/eR-KyPry)

8

Building with Arabic LLMs? 🛠️ New research just dropped a treasure trove of open-source resources: ✅ AraLLaMA: A 3x faster LLM ✅ HARNESS: Lightweight speech models ✅ MENAValues: A cultural alignment benchmark Get the full rundown in our latest digest: https://lnkd.in/dX6f5bj4

76

1 Comment

🚨 New Research on Large Language Models and Interpretability: 1. Their tendency to self-repair makes it difficult to trace why a particular output happened. 2. That breaks the chain of causal interpretability. 3. And without causal clarity, true explainability remains out of reach. 📄 In our latest paper, we propose GIM a method to tackle this head-on, unlocking more reliable interpretability in LLMs. 📎 Link to the paper: https://lnkd.in/dKXH4vfC Congratulations to Joakim Edin et al. Collaboration between Corti, Stanford University, Copenhagen University and LUT University.

68

1 Comment

AI is no longer just a technology story, it’s an economic power story. I share some thoughts on how AI is reshaping U.S. global influence and why Arab economies face a strategic moment: remain consumers of AI, or build the foundations to shape it.

19