Bozhen Peng

Back in May, I was invited by the local chapter of IEEE SSCS in Santa Clara to give a talk on how to use LLM and Gen AI for high-speed design. The video has been uploaded to YouTube. It is a good summary of how we are using LLM and AI for SI. https://lnkd.in/eUy2HDhB A few months later, Eric Schmidt gave a controversial talk on where he thought AI would progress in the next few years in a Stanford computer science class. He was a little too frank on telling students to steal TikTok IPs and why he thought Google would fail. https://lnkd.in/epMKkmvv Both talks seem to completely overlap on how to create domain knowledge dominance using advanced LLM. He gave examples on taking over TikTok and in chemical engineering, I was focusing on high-speed SI designs. But it can be applied to any other hardware design disciplines.  What we are pursuing, and he seems to suggest the same approach : 1) Use LLM with deep context window to scan through large volume of knowledge (text, video or sound), in my case for SI/PI we will be scanning large volumes of design specs for design domain knowledge 2) Use LLM agents to execute task based on the ingested knowledge (context to action), in my case, we will be initiating simulations based on design spec range and look for optimal solutions 3) Repeat the above with multiple chain of thoughts and store the observation to draw a conclusion, in my case, after scanning through the design specs and obtain optimal spacing for performance, generate routing and constraint rules 4) We also add an optimization step which is necessary for engineering  5) Improve the LLM agent performance using an adversarial agent to critique the LLM agent's performance 6) We also add using fine-tuning and transfer learning to future proof the domain specific models. I apologize both are long talks ~1 hour.  I hope this will be a good starting point to discuss the use of LLM and Gen AI for SI/PI. Feedback and comments are most welcome.

13

2 Comments

Hello fellow engineers & Pythonistas! 👋 I’m excited to share that I’ll be speaking about Pydantic and FastAPI at the upcoming Python Meetup on October 21 in Singapore! 🚀 Thanks to Martin Brochhaus for this opportunity! 🙏 https://lnkd.in/gYPbASS9 Let’s connect, learn, and innovate together! If you’re attending, I would love to hear your thoughts or any suggestion you might have. Looking forward to seeing you all there! #Python #FastAPI #Pydantic #PythonMeetup

50

3 Comments

NVIDIA IS BANG ON!!! JENSEN HUANG JUST MADE SOME REALLY COOL ANNOUNCEMENTS! https://lnkd.in/gZXh9qrQ *In the future a team of AIs will write software, and these chatbot AI developers will all be supervised by a super AI* ~ Nvidia CEO Jensen Huang He said it's been 60 years where no new technological revolution took place and said every great thing in computing took place after his birth! ✅ IBM S/360 ✅ Windows Pentium ✅ iPhone launch ✅ Internet Mobile Cloud He said this is the time for new age computing. Summing up the keynote for u- ☑ Stating the CEO math "the more you buy, the more you save.." he unveiled adding CPU to GPU, acclerating the speed unbelievably!!! ☑ NVIDIA CUDA libraries opens new markets. ☑ NVIDIA announces PANDAS-CUDF in GOOGLE collab. CUDA now achieves a virtual cycle, he says "If there are no developers that use ur platform than ofcourse there will be no users, no users means no installbase simply stating a 100x speeup translate to 98% savings". ☑ On March 18, 2024, NVIDIA launched the Earth-2 climate digital twin cloud platform, a pioneering solution to combat the staggering economic losses incurred from extreme weather events due to climate change, estimated at $140 billion. ☑ A new industrial revolution-introducing the AI Factory. ☑ Generative AI drives full stack reinvention-Software factory to AI factory. ☑ Stating CHATGPT several times, Nvidia seems to have decided that the graphics card industry was not worth bothering with and was moving to AI chips instead and becoming a major contributor to the AI revolution. ☑ He also introduced several Robots like last time and how physical AI is going to be a very important thing in the upcoming future. ☑ Stating "robots are gonna build robots" he adds, is gonna be a very important concept in robotics. And much more... Thanks for the read:) For expanded view on each of these and some deeper insights, follow the link below!

1

🌟 Excited to Announce My Summer Internship! 🌟 I am thrilled to share that I am interning this summer under the guidance of Professor Lin Zhong in the Efficient Computing Lab at Yale University, with mentorship from Zhiyao Ma. My project is creating automated tests for Hopter Embedded OS. Throughout this internship, I will be focusing on developing automated tests for Hopter’s kernel and integrating them with GitHub Actions. These tests will be crucial in ensuring code quality and stability by catching regressions early and facilitating smoother collaboration among developers. Key Highlights of My Project: - Setting up Virtualized Testing Environments: Using GitHub Actions and QEMU to create robust testing workflows. - Designing and Implementing Test Cases: From basic kernel operations to complex application interfaces and kernel internal modules. - Innovative Testing with Drones: Configuring a custom test server connected to a Crazyflie drone to test flight control applications. Technologies & Tools Involved: - GitHub Actions - Embedded Rust - QEMU Why This Project Matters: Automated testing is pivotal for maintaining the integrity and reliability of Hopter’s kernel. By implementing these tests, we aim to catch subtle errors early, provide peace of mind to code contributors and streamline the overall development process. I am eager to dive into this project and contribute to the innovative work being done at Yale's Efficient Computing Lab. Looking forward to an enriching learning experience and making a meaningful impact! #Internship #YaleUniversity #Yale #EfficientComputingLab #TechResearch #Research #EmbeddedSystems #Automation #GitHubActions #Rust #DroneTechnology #SoftwareTesting #FirstGenerationCollegeStudent #WomenInTech #ArabWomenInTech #ImmigrantStories #DiversityInTech #WomenWhoCode

24

4 Comments

Update on Headstarter Fellowship: 3 Project in 3 weeks week # 1 (7/22): Personal portfolio During the first week, I designed and developed my own personal portfolio which includes my personal projects, technology I am interested in working with and a simple landing page. live link: https://birongliu.dev Technology used: - ReactJS - TailwindCSS - TypeScript - framer-motion week # 2 (7/29): Pantry Tracker WebApp During the second week, I designed an web app that allow user to add items to their inventory and get AI to suggest foods based on the ingredients that is in the inventory live link: https://lnkd.in/ejWnjb6W source code: https://lnkd.in/eVMZSz3Y Improvement can be made: - Allow user to add ai suggested recipes to the list for future references - Allow user to have individual inventory instead of share across different users Technology Used: - NextJS - Firebase - TypeScript - TailwindCSS - Clerk - OpenRouter API week #3 (8/5): AI Support Chatbot For third week's project I have team up with Brian Martinez Flores and Faizan Khan to build an ai support chatbot that allow users to get suggestion from ai in regard about general finance advice such what form needed to file taxes and other aspects. live link: https://lnkd.in/efxtwzaB source code: https://lnkd.in/eTqwBxnK improvement can be made are: - use RAG (Retrieval Augmented Generation) to improve the ai response to users input based on authoritative knowledge such as feeding data into ai for matter understanding of market trend. - Store information using firebase such as user preferences and chat messages Technology used: NextJS TypeScript TailwindCSS Clerk OpenRouter API Yasin Ehsan 🚀 #computerscience

9

1 Comment

"I asked myself, 'Why is Japanese Quail trending?' after receiving an email from my school advertising the availability and sale of quail eggs. Additionally, I stumbled upon Bukola Oyebanji PhD, DVM post on LinkedIn about japanese quail. In the last academic session under monogastric and livestock production (APH 301), I learned about Japanese Quail as a type of poultry bird and I could recall their impressive features. Japanese Quail is a domesticated bird that thrives in small cages and is inexpensive to keep. While they're susceptible to common poultry diseases, they exhibit better resistance. Notably, they mature early and lay approximately 200-300 eggs per year. Their eggs are rich in high-quality protein, vitamins, minerals, and choline (essential for brain function and metabolism). Similarly, their meat has low fat content and is rich in iron, zinc, and selenium. Japanese Quail serves as an excellent substitute for chicken products. #JapaneseQuail #AnimalScience

6

How to Use Character Search FFXIV Ren Xu for Maximum Benefit This guide will focus on the character search ffxiv ren xu, exploring how to use this feature, its benefits, and some specific tips for finding the character Ren Xu. https://lnkd.in/diAvHKPm #charactersearchffxivrenxu #foryou #foryoupage #technology #article #tech #seo #onpageseo #website #blog #blogposting #posts #like #likesforlike #foryoupage #foryou #benefits #tech #guestpost #guestposting #blogging #blogpost #writing #like #share #follow #likes #innovation #management #HumanResources #digitalmarketing #technology #creativity #Future #futurism #Entrepreneurship #Careers #Markets #Startups #Marketing #SocialMedia #VentureCapita #SocialNetworking #LeanStartups #Economy #Economics #socialmedia #socialnetworking #venturecapital #motivation #sustainability #healthcare #education #design #sales #fundraising #construction

6

I recently completed the Multi-Agent Systems course from DeepLearning.AI taught by crewAI co-founder João (Joe) Moura crewAI offers an open-source version on their GitHub, allowing everyone to build their own multi-agent systems. It's already garnered 14.7k stars! The course is packed with fascinating insights. Here's a quick breakdown: Agent: Includes roles, goals, backstories, and optional tools. Task: Features a description, expected output, agent, and optional tools. Crew: Comprises agents, tasks, processes, and verbose information. You can run your tasks in sequence, hierarchically, or asynchronously, depending on your use case. I found it strikingly similar to Agile methodology, where we identify the right resources and write clear user stories and criteria. In the coming weeks, I'll be diving into the code and sharing my learnings. Stay tuned! GitHub: https://lnkd.in/gVs5mR3C #deeplearning.ai #DeepLearning #AI #MultiAgentSystems #OpenSource #MachineLearning #AgileMethodology #ArtificialIntelligence #CodeLearning #GitHub #CrewAI #DataScience #Programming #SoftwareDevelopment #Innovation #Coding #TechLearning #LLM #Generativeai #GenAI https://lnkd.in/gB2-djhh

7

CODE FOR ALL HACKATHON ANNOUNCEMENT!!! Interested in participating in hackathons, but they are always so far and competitive? Code for All QC has got you covered! We will be hosting our first ever beginner-friendly hackathon for Queens College and CUNY students on November 8th! This will be an in-person hackathon on the Queens College spanning from November 8th to the 9th, where you will be able to compete with your fellow students for a chance to win! Dates November 8th, 9:30 AM - November 9th, 1 PM Opening Ceremony - 9:30 AM 11/8 Hacking Begins - 10:00 AM 11/8 Hacking Ends - 10:00 AM 11/9 Presentations - 11:00 AM 11/9 Announcing Winners - 12:30 - 1:00 PM 11/9 Eligibility Must be a Queens College or CUNY Student! 17+ years of age Must be in a team of 2+ members. A maximum of 4 members are allowed in a team. Last-minute teams may be formed day of the event Project and Submission Requirements Your project need not be fully functioning. If it's not functioning, you may submit another version of the project (design, prototype) Your project need not be deployed, but extra brownie points if it is! Your project must not have been worked on before the start of the event. We will be checking your git and file history. Prizes Certificates will be given to winners of certain tracks. Best Overall Hack Runner-Up Hack Best Design Best Beginner Hack Best Health Track Hack Best Environment Track Hack Best Education Track Hack Best Use of AI Judging Criteria and Winner Selection To be considered to win, you must present your project in front of judges! Projects are judged based on creativity, impact, design, and more. We need many of you to show out so we can host bigger hackathons in the future! Register yourself here and gain more info: https://lnkd.in/eFSzrimp Team registration: https://lnkd.in/e4F6uRSZ (teams don't need to be formed yet to register) Room numbers to be announced

88

6 Comments

💥💥💥 EoRA: Training-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation Shih-Yang Liu, Huck Yang, Chein-Yi Wang, Nai Chit Fung, Hongxu Yin, Charbel Sakr, Saurav Muralidharan, Kwang-Ting Cheng, Jan Kautz, Yu-Chiang Frank Wang, Pavlo Molchanov, Min-Hung Chen Abstract In this work, we re-formulate the model compression problem into the customized compensation problem: Given a compressed model, we aim to introduce residual low-rank paths to compensate for compression errors under customized requirements from users (e.g., tasks, compression ratios), resulting in greater flexibility in adjusting overall capacity without being constrained by specific compression formats. However, naively applying SVD to derive residual paths causes suboptimal utilization of the low-rank representation capacity. Instead, we propose Training-free Eigenspace Low-Rank Approximation (EoRA), a method that directly minimizes compression-induced errors without requiring gradient-based training, achieving fast optimization in minutes using a small amount of calibration data. EoRA projects compression errors into the eigenspace of input activations, leveraging eigenvalues to effectively prioritize the reconstruction of high-importance error components. Moreover, EoRA can be seamlessly integrated with fine-tuning and quantization to further improve effectiveness and efficiency. EoRA consistently outperforms previous methods in compensating errors for compressed LLaMA2/3 models on various tasks, such as language generation, commonsense reasoning, and math reasoning tasks (e.g., 31.31%/12.88% and 9.69% improvements on ARC-Easy/ARC-Challenge and MathQA when compensating LLaMA3-8B that is quantized to 4-bit and pruned to 2:4 sparsity). EoRA offers a scalable, training-free solution to compensate for compression errors, making it a powerful tool to deploy LLMs in various capacity and efficiency requirements. 👉 https://lnkd.in/dwwJv7PZ #machinelearning

2

🇯🇵 Microsoft Launches AI and Robotics R&D Hub in Tokyo: Bridging Tech and Expertise Microsoft has taken a significant step in advancing AI and robotics by opening a new research and development center in Tokyo’s Minato Ward. This initiative is designed to combine Japan’s robotics expertise with cutting-edge AI technology for real-world applications. Key Collaborations 🤝 Partnering with leading organizations like Kawasaki Heavy Industries and Nissan Motor. 🎓 Collaborating with local universities to foster innovation and talent. Why This Matters • Tackling Japan’s aging population and shrinking workforce by driving AI-driven automation and robotics solutions. • Offering AI upskilling programs to empower individuals and industries with future-ready skills. • Expanding data center capacity and AI research to support Japan’s growing tech ecosystem. This initiative aligns with Microsoft’s global vision to integrate technology for societal challenges while strengthening its footprint in AI innovation. 🌏 Could this be the blueprint for solving global workforce challenges with AI and robotics? https://lnkd.in/gCHgiAQu #AI #Robotics #Innovation #Microsoft #Japan #FutureOfWork #TechForGood

4

Nvidia CEO Jen-Hsun Huang Sold Shares Worth $27M – No Confidence or Just Profit Booking? 🤔💰 Following up on my previous analysis of Nvidia, I'm excited to share an updated report on the company. Since the last one-pager, DCF, and technical analysis, Nvidia's stock not only fell to the estimated number but has also shown both positive and negative signs of strength. Nvidia remains in the spotlight due to the AI hype, with its GPUs recognized as top performers for high-performance tasks. However, these GPUs face competition from AMD's chiplet GPUs, which may offer similar performance at a lower cost. The stock has experienced a significant drop of 22.3% from its all-time high. The updated analysis highlights this decline and notes that the stock is now below its 50 SMA, indicating short-term bearishness. Notably, Nvidia's founder and other insiders have sold a substantial portion of their holdings, with the CEO alone selling over $27 million in company stock. This could be a simple profit booking due to the expectation of an increase in capital gains. While the report captures these transactions, it does not account for Nancy Pelosi's recent purchase of 10,000 shares, which occurred post-report. Vertisuite Investment & Consultancy Club (VICC) had projected a potential fall to around $100 in the last report. The Biden administration's new restrictions on exporting high-tech chips to China could serve as a boon to Nvidia. The report delves into these dynamics, balancing the positives and negatives and providing an updated technical perspective on Nvidia. As always, it's crucial to stay informed and consider multiple factors when analyzing market movements. #stocks #stockmarket #nvidia #techincalanalysis #dcf

26

5 Comments

Today, I am excited to have presented my research project on Shop the Look: Discover Your Style With AI and Visual Search for Fashion (https://lnkd.in/gvdPsm5s) on the Industrial Day at the University of Massachusetts Dartmouth! 🎉 My project focuses on building a custom YOLOv5 model to detect fashion accessories like shirts, pants, shoes, handbags, and sunglasses. The goal is to streamline product recommendations and transform visual inspiration into actionable shopping options, much like Amazon's and Pinterest's "Shop the Look" feature. The pipeline contains the following steps:👇 1️⃣ Data Collection • Scraped 100 images per class from Pinterest, focusing on various clothing accessories. 2️⃣ Data Annotation • Annotated images using Label Studio, creating bounding boxes around objects and exporting them in YOLO format. 3️⃣ Model Training • Trained a YOLOv5 model using YOLOv5-large weights, fine-tuned with PyTorch to match custom classes. 4️⃣ Feature Extraction • Used a pre-trained ResNet-50 model to extract feature embeddings from the input images for further analysis and recommendation. 5️⃣ Running Inference • Applied the trained model to new images, detecting clothing accessories using OpenCV-DNN, and providing similar fashion recommendations based on the detected objects. 6️⃣ Visualizing the Results • Displayed detected objects with bounding boxes and labels to validate the model's accuracy, while offering recommendations for similar fashion items. GitHub repository for the project: https://lnkd.in/gNKuteeq For more details, check out Kunlong Gu's insightful Pinterest blog: https://lnkd.in/gN87digb A big thank you to Kunlong Gu for sharing this fantastic resource! This project draws inspiration from both Pinterest Engineering and Amazon Science. I am currently exploring Spotify's ANNOY (Approximate Nearest Neighbors Oh Yeah) for generating similar recommendations on an article in a low-latency runtime environment. Stay tuned. I will share the updates soon! 💫 🙌 References: • Approximate Nearest Neighbors by Spotify: https://lnkd.in/gPMAq6NM • Vector Similarity Search by Google: https://lnkd.in/gfY3UGC5 • Efficient Similarity Search by Facebook: https://lnkd.in/gn_EkHa6 It was a great opportunity to connect with fellow students and share insights. #YOLOv5 #ResNet50 #DataAnnotation #ComputerVision #FashionRecommendation #IndustrialDay

88

6 Comments

Check out Prof Jason Corso’s blog, “From the Kitchen to the Lab: Why Cooking Became AI’s Favorite Dish” - https://lnkd.in/g85W8XRV Cooking, with its rich visual complexity and procedural, goal-oriented nature, offers a fascinating domain for data-driven research. Through a review of over a decade's worth of cooking-oriented datasets and the innovations they've spurred, this essay delves into the meat of why and how the culinary world has gained a foothold as a key research domain for CV/ML. #computervision #ai #artificialintelligence #machinevision #machinelearning #datascience

2

一邊聽著黃仁勳的演講,一邊產生一個疑問: 我會喜歡對著搞不清楚是AI還是真人的影像嗎?還是對著機器面板講話? My question raised from Jensen Huang, CEO of nVidia's speech: I have to talk with AI; however, do I want to face the machine with human face or just the display?

1