David Luebke

Tired of marching cubes producing skinny triangles and losing sharp geometric details? The NVIDIA Kaolin library for #3d #deeplearning and #AIresearch now includes FlexiCubes, a method for extracting meshes from scalar fields designated for gradient-based mesh optimization. Check out this tutorial: https://lnkd.in/g2zQP3aX

42

Our paper, "DRAMScope: Uncovering DRAM Microarchitecture and Characteristics by Issuing Memory Commands," was accepted at ISCA 2024! We had an intriguing journey to figure out the sizes of DRAM MATs, common pitfalls regarding address and data mapping in the computer architecture community, coupled-row activation, and RowHammer/RowPress characteristics induced by the 6F^2 DRAM cell structure, to list a few, by issuing ordinary commands to DRAM. We look forward to presenting our paper in Buenos Aires, Argentina. I am really proud of my students and colleagues, without whom I could not imagine to publish this work in a one-and-done manner. Pdf: https://lnkd.in/gnnQS_XA Authors: Hwayong Nam, Seungmin Baek, Minbok Wi, Michael Jaemin Kim, Jaehyun Park, Chihun Song, Nam Sung Kim, and Jung Ho Ahn Abstract: The demand for precise information on DRAM microarchitectures and error characteristics has surged, driven by the need to explore processing in memory, enhance reliability, and mitigate security vulnerability. Nonetheless, DRAM manufacturers have disclosed only a limited amount of information, making it difficult to find specific information on their DRAM microarchitectures. This paper addresses this gap by presenting more rigorous findings on the microarchitectures of commodity DRAM chips and their impacts on the characteristics of activate-induced bitflips (AIBs), such as RowHammer and RowPress. The previous studies have also attempted to understand the DRAM microarchitectures and associated behaviors, but we have found some of their results to be misled by inaccurate address mapping and internal data swizzling or lack of a deeper understanding of the modern DRAM cell structure. For accurate and efficient reverse engineering, we use three tools: AIBs, retention time test, and RowCopy, which can be cross-validated. With these three tools, we first take a macroscopic view of modern DRAM chips to uncover the size, structure, and operation of their subarrays, memory array tiles (MATs), and rows. Then, we analyze AIB characteristics based on the microscopic view of the DRAM microarchitecture, such as 6F^2 cell layout, through which we rectify misunderstandings regarding AIBs and discover a new data pattern that accelerates AIBs. Lastly, based on our findings at both macroscopic and microscopic levels, we identify previously unknown AIB vulnerabilities and propose a simple yet effective protection solution.

152

3 Comments

We invite research submissions on Quantum Computing and Reinforcement Learning (QCRL) at IEEE Quantum Week 2024 (Montreal, Sep. 15-20, 2024). https://2024.qcrl.io/ Scope and Topics: Topics relevant to this session include, but are not limited to: - Theoretical studies of quantum reinforcement learning algorithms - Challenges in quantum reinforcement learning in the NISQ era - Quantum reinforcement learning under the influence of quantum device noise QRL in the context of trustworthy ML (e.g., adversarial attacks, federated learning) - Quantum reinforcement learning for scientific discovery, commercial, and industrial applications - Benchmarking QRL algorithms - Reinforcement learning for quantum control, quantum error correction, and quantum error mitigation - Reinforcement learning for quantum architecture search, optimization, and program synthesis - Hybrid RL-QC systems design Important Deadlines: - Jul 10, 2024: Abstract Due Date - Jul 15, 2024: Paper Submission Due Date - Jul 28, 2024: Notification of Paper Acceptance - Aug 5, 2024: Camera-ready Papers Due #quantum #machinelearning #reinforcementlearning

30

We are pleased to share the graphical abstract "A Reconfigurable Near-Sensor Processor for Anomaly Detection in Limb Prostheses". 📝 Authored by: Jiayu Huang; Zikai Zhu; Peng Su; Dejiu Chen; Li-Rong Zheng; Zhuo Zou This paper presents a reconfigurable near-sensor anomaly detection processor to real-time monitor the potential anomalous behaviors of amputees with limb prostheses, which comprises a reconfigurable Variational Autoencoder (VAE), a scalable Self-Organizing Map (SOM) Array, and a window-size-adjustable Markov Chain, achieving an energy efficiency of 43.84 nJwithin 24.22 μsper detection. 🔗 Read more on IEEE Xplore: https://loom.ly/BaNKYYI We invite you to watch the video and pause as needed to examine the research in detail. #IEEE #IEEECASS #TBioCAS #Biomedical #CircuitsandSystems #graphicalabstract

18

The next honorable mention for the 2023 IEEE Transactions on #Robotics King-Sun Fu Memorial Best Paper Award is a paper we shared several months ago. Paper title: Kinegami: Algorithmic Design of Compliant Kinematic Chains From Tubular Origami Authors: Wei-Hsi Chen , Woohyeok Yang, Lucien Peach, Daniel E. Koditschek GRASP, and Cynthia Sung Paper link: https://lnkd.in/ggRQ7cQf Video caption: The supplemental video presents the introduction, fabrication, and application of a catalog of tubular origami modules and their compositions. It also shows an animated description of the "Kinegami" algorithm and includes a demonstration of an actuated Kinegami catapult. #RobotKinematics #Actuators #OrigamiRobot

164

2 Comments

I'm glad to share our new paper, which reports on the development of an e-tongue for sensitive monitoring of neurotransmitters using four different #MXenes as sensing units. Note that the first two authors were visiting scientists from Brazil and Azerbaijan at A.J. Drexel Nanomaterials Institute. Without them, this work would not be done, and we would not even start it. This shows the importance of bringing creative international visitors who implement their original ideas. Increasing restrictions will badly hurt the scientific productivity of US research labs. #research #science #MXene, #etongue #sensors #collaboration #visitingscientists

166

7 Comments

I'm excited to share about the good work from my colleagues and collaborators from Taiwan Tshing Hua Univ https://lnkd.in/g2dHAAic Maximum Entropy Reinforcement Learning via Energy-Based Normalizing Flow Abstract Existing Maximum-Entropy (MaxEnt) Reinforcement Learning (RL) methods for continuous action spaces are typically formulated based on actor-critic frameworks and optimized through alternating steps of policy evaluation and policy improvement. In the policy evaluation steps, the critic is updated to capture the soft Q-function. In the policy improvement steps, the actor is adjusted in accordance with the updated soft Q-function. In this paper, we introduce a new MaxEnt RL framework modeled using Energy-Based Normalizing Flows (EBFlow). This framework integrates the policy evaluation steps and the policy improvement steps, resulting in a single objective training process. Our method enables the calculation of the soft value function used in the policy evaluation target without Monte Carlo approximation. Moreover, this design supports the modeling of multi-modal action distributions while facilitating efficient action sampling. To evaluate the performance of our method, we conducted experiments on the MuJoCo benchmark suite and a number of high-dimensional robotic tasks simulated by Omniverse Isaac Gym. The evaluation results demonstrate that our method achieves superior performance compared to widely-adopted representative baselines.

52

1 Comment

I am very excited to share our latest paper, "Emergence of Multiphoton Quantum Coherence by Light Propagation," now on arXiv! We’ve discovered that scattering thermal multiphoton wavepackets in free space can modify quantum coherence, leading to light fields with unique quantum properties. This could revolutionize various quantum technologies by enabling the creation of sub-shot-noise quantum systems. Check it out here: https://lnkd.in/gV89mhRr #QuantumPhysics #Photonics

25

2 Comments

Using a machine learning model and EEGs, the ECE Capstone team of Jioh Kim, Nick Short, Paula Perdomo, and Steven Zhang reconstructed the mental image formed when a subject was shown a picture — tech with the potential to help those who are conscious but unable to communicate. Perhaps even a newborn? That would be cool! You can read more about the ECE Capstone Design Fair here: uoft.me/avG. #UofT #UofTEngineering

12

1 Comment

Did you know that - when making video games or animated movies - people are basically creating a 3D world and then simulating how the light reflects from objects and hits your eye? How do they do that, you ask? Good old Monte Carlo is the answer. And what makes one Monte Carlo better than the next Monte Carlo? Better sampling techniques. So what if I told you, that we have come up with a way to optimize the noise samples to make Monte Carlo great (again)? Wouldn't that be a really neat way to create better experiences and higher quality rendering. It sure is! This is really William Donnelly and Alan Wolfe's baby. I tried a trillian ways to train a ML model to do the same thing that random search achieves and I failed a trillion times. But well, at least I have learned a thing or two about rendering.

25

1 Comment

We will be presenting our recent IEEE RA-L publication tomorrow (5/16) at #ICRA2024! (A collaboration with Shin Hirofumi from Honda R&D Co, Ltd) “Reinforcement Learning with Energy-Exchange Dynamics for Spring-Loaded Biped Robot Walking” Please join our oral session at the “Humanoid and Biped Locomotion Session (10:30, AX-F204)”, and our interactive session at booth 16.01 from 13:30 to 15:00. Our research uses model-based reinforcement learning to achieve “on-site dynamics learning” and “real-time control planning” for a spring-loaded bipedal robot. We express the robot’s proprioception as an energy formulation and use model-based RL to learn the reduced-order state-transition dynamics. Simulation and hardware experiment results demonstrate that our approach can quickly learn the dynamics of a 4-DoF spring-loaded bipedal robot and accomplish walking tasks in just minutes of sampling.

12

Check out the latest research on high-performance graph processing from a hardware/software co-design perspective! This paper dives deep into innovative strategies to optimize computational efficiency. https://lnkd.in/gUXumc5x 廖小飞 Hai Jin Long Zheng Zhiyuan Shao

Nemotron-4 340B is a huge release by NVIDIA! Why? The Nemotron-4 340B instruct model lets you generate high-quality data and then the reward model (also released) can filter out data on several attributes. It's often not enough just to have a good instruct model for generating high-quality synthetic data. You also need to filter the generated data on a set of requirements. It's common to use the same model to grade the outputs (LLM-as-Judge) but the reward model (Reward-Model-as-Judge), which is often not released, is ideal for this. Also, check out in the paper how Nemotron-4-340B compares with GPT-4-1106-preview when evaluated by humans on several tasks. Seems that GPT-4 is a lot better at rewrite and extraction but Nemotron-4-340B compared nicely on other tasks. Multi-turn chat has a high score! Something else that caught my attention is that the majority of preference data used for alignment are synthetic. The results show that Nemotron-4 340B is a strong model. Check out those MMLU, GSM8K, and Arena Hard numbers in the paper. GPT-4 is still ahead but the gap keeps closing. What's Meta's move now? Are we going to see a similar type of release? Could be a huge moment for the field and open LLMs. Nice release from NVIDIA. They even released a preference dataset (link in the replies). Well done to the team! Blog: https://lnkd.in/eSRbajCw Paper: https://lnkd.in/easmAUz7 Dataset: https://lnkd.in/ekGFsEXj

548

2 Comments

Read #NewPaper “Assessing Cardiac Functions of Zebrafish from Echocardiography Using Deep Learning” by Mao-Hsiang Huang et al. See more details at: https://lnkd.in/gQy4b_Wd #deeplearning #heartdisease

2

With a small amount of architectural extensions, our recently accepted paper "M^3XU: Achieving High-Precision and Complex Matrix Multiplication with Low-Precision MXUs" can let low-precision AI/ML-oriented matrix processors capable of performing IEEE 754 32-bit floating point, 64-bit double-precision operations and even more, native support for complex number operations! Programmers do not need to worry about changing their code or new accuracy issues -- the resulting architecture generates exactly the same outcome as standard 32-bit/64-bit floating numbers, but at least doubled performance! M^3XU will be published in SC24 in November and we will release the framework later! This work is done together with my students Dongho Ha Yunan Zhang at UCR, my collaborators, Chen-Chien Kao (UMich), Won Woo Ro (Yonsei University) and Christopher Hughes (Intel).

98