Mat Gilbert, PhD

Today @ AGI House SF’s Launch-a-thon, I’m launching Converge: An open-source AI-Human developed programming language for Neuromorphic Computing on Intel Corporation’s Hala Point system built atop their Loihi 2 experimental chip. Neuromorphic computation architecture is designed to mimic the parallel processing if the human brain. This offers an interesting precedent when it comes to not only designing the language framework and libraries itself, but the methodology to adversarially attack such a system, which I hypothesize is to fashion perturbations modeled off of / inspired by disordered neurological phenomena such as schizophrenia, Alzheimer’s, ADHD, mania / bipolar disorder, psychosis, etc. Launching and making the first commit to the repo at 6 PM PT: https://lnkd.in/eS6566f9 black dream research

13

3 Comments

In 2016, as chair of the PBS Engineering Technology Advisory Committee (ETAC), I introduced the potential of photonic circuit switches to a broad public media ETAC Member stations through a PBS Technews column. Photonic circuit switches, utilizing photons instead of electrons, significantly reduce energy consumption. By miniaturizing light-guiding components onto silicon chips, they provide a powerful solution for maximizing expensive resources. This technology holds particular promise for applications demanding high bandwidth, such as remote sports production and telematic performance. Today, the demand for high-quality, big-screen content underscores the critical role of optical communication in robust media networks. Thankfully, advancements in integrated switches are delivering real-world benefits, paving the way for a more sustainable and efficient future of connectivity.

11

MathWorks welcomes Cadence Design Systems #Tensilica #DSP into the Connections program for an integration that enables rapid porting of #MATLAB and #Simulink models, including #PIL (processor-in-the-loop) capability as well as verification and testing of generated code.

12

I’ve personally witnessed at least three “AI springs” and two “AI winters”. I wanted to add my perspective on the impact of this latest AI evolution on software engineering - in particular system design. Software development is a complex, highly-skilled knowledge job that demands more than just coding skills—it requires innovation, abstract reasoning, and creative problem-solving, capabilities where human intelligence excels and AI often falls short. However, AI can support some of the tasks in system design to make devs more productive, such as helping engineers understand system requirements, assess the impact of their decisions, proactively suggest system solutions or optimizations, etc. Read the full article here 👇

2

New Case Study Alert!  🎉 We’re excited to share our latest case study that dives into innovative meshing techniques. 🛫 This study explores how an enhanced boundary layer mesh generated with #FidelityPointwise by Cadence, and applied to a wing featuring a narrow slot for mass injection, can enable precise prediction of flow separation patterns. 🔍 Whether you’re an aerospace engineer, CFD enthusiast, or simply curious about advancements in meshing, this case study provides valuable insights into: •⁠ ⁠the impact of mesh refinement techniques •⁠ ⁠accurate modeling of complex flow scenarios •⁠ ⁠practical applications for enhancing mesh control. 📖 Read the full case study here: https://lnkd.in/diACUF_j We’d love to hear your thoughts and feedback! Let’s discuss how these findings might apply to your challenges or projects. #AerospaceEngineering #CFD #meshing

12

One of NVIDIA's superpowers is our ability to build a thriving ecosystem of partners. We develop our ecosystem strategies by industry and even create sub-ecosystems specific by application. One example of an application-specific partnership is that between NVIDIA Healthcare and MathWorks. Together the 2 organizations are empowering medical device engineers to leverage AI more effectively, speeding up the development and deployment of real-time, software-defined medical workflows. By incorporating #MATLAB’s algorithms directly into the Holoscan platform, we're enabling a seamless transition from development to deployment, ensuring that AI applications can be brought into clinical settings with ease. Thank you to Marc Edgar, Akhilesh Mishra, Mostafa Toloui and their teams for making the integration a reality. #nvidiahealthcare #nvidia #acceleratedcomputing #medtech #medicaldevices #medicalimaging #digitalsurgery #roboticsurgery #aiml #artificialintelligence #softwaredefined

34

I’d like to implement the concept of brains for an AI code reviewer. I think this would minimize the context window size needed, and therefore maximize the utility of asking claude/llama/chatGPT to “review this repository” https://lnkd.in/g2U_qDdj https://lnkd.in/g6n_9kpi

1

Came across this intriguing article in the New York Times about OpenAI's latest development - a deepfake detection tool with impressive accuracy. https://lnkd.in/gazYaZha While this is a significant step forward in combating misinformation, it also sparks some interesting questions. Here are my thoughts: Deeper Fakes, Deeper Detection: This AI-powered detection is a powerful defense mechanism. As deepfakes become more sophisticated, having reliable ways to identify them becomes ever more crucial. The Cat-and-Mouse Game: The creators of deepfakes will likely adapt their methods to bypass this new detector. This necessitates continuous improvement and adaptation of detection algorithms, making it an ongoing battle. Ethical Considerations: Widespread use of deepfake detection raises ethical concerns. Biases in the algorithms could lead to misidentification of legitimate content. We need open discussions and transparent development to ensure fairness and avoid misuse. Overall, OpenAI's work is a positive development. However, it's just one piece of the puzzle. We need to have a multi-pronged approach that combines technological advancements with media literacy initiatives to address the deepfake challenge effectively. 𝗟𝗲𝘁'𝘀 𝗴𝗲𝘁 𝗮 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝗴𝗼𝗶𝗻𝗴! 𝗪𝗵𝗮𝘁 𝗮𝗿𝗲 𝘆𝗼𝘂𝗿 𝘁𝗵𝗼𝘂𝗴𝗵𝘁𝘀 𝗼𝗻 𝗱𝗲𝗲𝗽𝗳𝗮𝗸𝗲𝘀 𝗮𝗻𝗱 𝘁𝗵𝗲 𝗳𝗶𝗴𝗵𝘁 𝗮𝗴𝗮𝗶𝗻𝘀𝘁 𝗺𝗶𝘀𝗶𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻? #deepfakes #AI #technology #opendiscussion #AI #machinelearning #deeplearning #technology #innovation #opendata #opendiscussion #misinformation #fakenews #deepfakes #security #privacy #ethics #bias #algorithmicbias #responsibleAI #media #medialiteracy #futureofwork #futureOfWork #techforgood #technologicaladvancement #communication #communicationstrategy #informationsecurity #cybersecurity #NYTimes #OpenAI

5

Last week we ran a webinar on implementing maths in FPGA. One of the attendees asked the question what the difference in implementation size would be between fixed and floating point. I thought it would make for a pretty good blog, so this week we are implementing the same algorithm in floating and fixed point maths and seeing what the differences are. #fpga #embeddedsystems #engineering #embeddedsoftware #engineering https://lnkd.in/eR4TUYth

163

16 Comments

🔍 Learn About Covert Timing Channels with Kathará! 🔍 Ever wondered what a Covert Timing Channel is? Now you can learn it through practice using Kathará! Mario Cuomo has shared with us an interesting network scenario demonstrating how to implement a simple Covert Timing Channel using Python. Check it out here: https://lnkd.in/dmy4WH2k This scenario is part of Mario's master thesis project, which also led to a publication in the Advances in Science, Technology and Engineering Systems Journal titled: “Detecting CTC Attack in IoMT Communications using Deep Learning Approach.” Read the paper here: https://lnkd.in/dYNKnykn Many thanks to Mario for sharing this valuable resource 😍 #NetworkEmulation #CovertTimingChannel #CyberSecurity #Education #Kathará #DeepLearning #CommunityLabs

8

New FFT development notes: The ideas I've developed for a new FFT have been tested on a GPU for a number of different 1D vector lengths. The speedups obtained were as follows: Length 99, speedup of 44.5x; length 199, speedup of 105.5 x; length 499, speedup of 207.7 x, and length 1023, speedup of 5.6x. Results of other speedups will be announced soon for the 2D FFT.

12

3 Comments

https://lnkd.in/eXbCsAnD “He who controls the GPUs, controls the universe.” Access to GPUs, and in particular those made by Nvidia, the leading supplier, is vital for any company that wants to be taken seriously in #ai. Nvidia’s dominance has driven its market value to more than $2trn. 'GPUs do the computational heavy-lifting needed to train & operate large #ai models. Yet, oddly, this is not what they were designed for. The initials stand for “graphics processing unit”, because such chips were originally designed to process video-game graphics. It turned out that, fortunately for Nvidia, they could be repurposed for #ai workloads. 'Might it be better to design specialist ai chips from scratch? That is what many companies, small & large, are now doing in a bid to topple Nvidia. Dedicated ai chips promise to make building and running ai models faster, cheaper or both. Any firm that can mount a credible threat will have no shortage of customers, who dislike its lofty prices and limited supplies. 'GPUs contain several thousand processing engines, or “cores”, which let them run thousands of versions of the same simple task at the same time. Running #ai models similarly involves running lots of copies of the same task in parallel. Figuring out how to rewrite ai code to run on GPUs was one of the factors that triggered the current ai boom. 'Yet GPUs have their limitations, particularly when it comes to the speed with which data can be shuffled on and off them... When training very large #ai models, some GPu cores may be idle as much as half of the time. Andrew Feldman, the boss of Cerebras, a startup based in Sunnyvale, CA, likens it to the gridlock in a grocery store on the day before Thanksgiving. “Everybody’s in a queue, so there are blockages in the parking lot, there are blockages in the aisles, blockages at the checkout. That’s exactly what’s happening with a gpu.” 'Cerebras’s response is to put 900,000 cores, plus lots of memory, onto a single, enormous chip, to reduce the complexity of connecting up multiple chips & piping data between them. 'Groq, a startup from Mountain View, CA, is taking a different approach. Its ai chips, called language processing units (lpus), are optimised to run #llms particularly quickly. 'Yet another approach is that taken by MatX, also based in CA. GPUs contain features & circuitry that provide flexibility for graphics, but are not needed for llms. The firm is working on getting rid of such unnecessary cruft.

14

3 Comments

📣MathWorks welcomes Texas Instruments to its Connections Program for yet another integration, the #AM26X family of controlCARD and LaunchPad boards, enabling automatic build, load, and run algorithms on AM26x devices using #Simulink models. Texas Instruments provides analog and #embedded semiconductors that are the essential building blocks of electronic systems. The Arm-based AM26x Microcontrollers are built to meet the complex real-time processing and control needs of next generation industrial and automotive embedded projects. The MathWorks Connections Program includes commercially offered products and services that complement #MATLAB and Simulink. https://spr.ly/6041v8sjB

3

Working with ADC and DACs is critical for FPGA developers. Last week I ran a webinar on how to work with ADCs and DACs. For this webinar we looked at the XADC and Delta Sigma / PWM DACs. As always the examples are provided on my GitHub which is linked with the video. https://lnkd.in/eyvmhtPA #fpga #electronics #embeddedsystem #embeddedsoftware #engineering

138

4 Comments

"This XXXX is really hard!" was a comment made by a peer recently regarding working toward 112/224G successful design work. Several key elements in the presentation will show how to make SI crux issues easier for high confidence design, pitfalls to not fall into. We introduce simple but fresh new perspectives on macro versus micro signal integrity (the team has other terminology for this), issues with 3D anisotropic material ID extraction, how fabrication tolerance impacts your SI in ways you may not have thought of.

23

5 Comments

📚 New Book Alert: Fundamentals of Radar Signal Processing 📡 Engineers and students, don't miss out on the latest edition of "Fundamentals of Radar Signal Processing" by Mark A. Richards. This essential resource is packed with comprehensive insights and practical techniques for mastering radar signal processing. 🔍 Book Highlights: >> In-Depth Coverage: Explore the fundamentals and advanced concepts of radar signal processing. >> Practical Techniques: Learn practical methods and algorithms used in real-world radar systems. >> Updated Content: Benefit from the latest advancements and updates in the field. >> Examples and Case Studies: Understand complex concepts through detailed examples and case studies. 💡 Why Read This Book? Stay ahead in your field by mastering the principles and applications of radar signal processing. Perfect for engineers, researchers, and students looking to deepen their knowledge and enhance their skills. 👉 Explore the Book: https://spr.ly/6047cXjRf #RadarSignalProcessing #Engineering #MATLAB #Simulink #BookAlert #Innovation #Education #MathWorks

3

⭐ I’m excited to announce our new blog post⭐ Intel Labs and Hugging Face introduce dynamic speculative decoding to 🤗 Transformers: a new trick that accelerate text generation by up to 2.7x It is so consistent that it is now the default in 🤗 Transformers when you trigger assisted generation/speculative decoding. Read the blogpost here: https://lnkd.in/d262VJbu How does it work? With speculative decoding, we split the generative process into two stages: 1️⃣ A small, but less accurate draft / assistant model generates a sequence of tokens 2️⃣ The target model applies parallelised verification over the tokens from the draft model This allows the target model to produce multiple tokens in a single forward pass and thus accelerate decoding. The approach relies on a concept known as speculation lookahead (SL), which is essentially the number of tokens generated by the draft model during each iteration. Now, SL is usually a static value or determined based on heuristics - in both cases this leaves a lot of performance on the table The trick behind dynamic speculative decoding is to dynamically adjust the number of draft tokens generated *per iteration*. This strikes a balance between generating too few draft tokens (which would forgo potential speed gains) and too many (which would result in tokens being discarded by the target model and reducing the speedup gain)." Happy decoding! Thanks to the awesome collaborators: Jonathan Mamou, João Gante, Lewis Tunstall, Daniel Korat, Nadav Timor, Moshe Wasserblat

68

4 Comments

Principal Application Engineer Philipp Jacobsohn and Director of FPGA Design Sunil Kumar have written a #whitepaper about porting #ASIC IP cores to #FPGA. If you’ve ever considered undertaking this sort of #engineering work, the paper offers a high-level blueprint of the many things you’ll need to consider — and the benefits that await! Score a free download of the white paper here: https://hubs.la/Q02_tTDx0

52