Graphcore’s Post

I am excited to share that, as part of #LASCAS2025 [1], Cristina Meinhardt, Guilherme Paim, and myself are organizing a special session on Approximate and Transprecision Computing: the Quest of More with Less. You are invited to submit your work! We invite original contributions, theoretical or applied, that include but are not limited to the following topics: - Design automation of approximate architectures - Design of approximate reconfigurable architectures - Error resilient Near-Threshold Computing - Approximation and Transprecision for Deep Learning applications - Approximation techniques for emerging processor and memory technologies - Approximation-induced error modeling and propagation - Approximation and Transprecision in edge computing applications and reconfigurable computing - Approximation and Transprecision in HPC and embedded systems - Architectural support for approximation and transprecision - Cross-layer approximate computing - Methods for monitoring and controlling approximation quality - Modeling, specification, and verification of approximate circuits and systems - Safety and reliability applications of approximate computing - Security in the context of approximation - Hardware/software co-design of approximate and transprecision systems - Dependability of approximate circuits and systems - Software-based fault-tolerant technique for approximate computing - Test and fault tolerance of approximate and transprecision systems [1] https://lnkd.in/eu_peBQz

This week, this dedicated group in NRIS is coming together to kick off the preparations for the production setting of our new supercomputer acquisition, currently known as "A2". 😍 👇   In an age where research in an increasing number of fields is data-driven, the demand for computing power is almost insatiable. Our new machine, a Cray Supercomputing EX system from Hewlett Packard Enterprise, is equipped with 304 of the most advanced #GPUs on the market today (NVIDIA GH200 96 GB). These GPUs not only have enormous computing capacity but are also compatible with a wide range of software and programming languages used in data-driven research, making them a popular choice among researchers, especially in #AI. 🚀   One of the key tasks of the group will be to ensure the best possible deployment of this substantial GPU power available to researchers in Norway next year. Another important task is to ensure a tightly integrated connection with the national Norwegian Infrastructure for Research Data (NIRD) and to ensure maximum utilisation of what will be Norway's most powerful supercomputing system ever. Let us see if the capacity will be enough for Norway’s ever-increasing AI needs! ⚙ 💪     Our tight collaboration in NRIS puts us in a unique position to tap into expert competence from the four collaborating universities (Universitetet i Bergen (UiB), Universitetet i Oslo (UiO), UiT Norges arktiske universitet and Norges teknisk-naturvitenskapelige universitet (NTNU)) and from Sigma2.    We are cheering on the working group and wish them the best of luck! 🙌 👏 Dhanya Pushpadas, Steinar Brattøy Gundersen, Truls Mathiassen, Stein Inge Knarbakk, Einar Næss Jensen, Dan Jonsson, Jenny Amundsen Ask and Ole Saastad

“In a normal computer vision system, you create a signal, then you have to transfer the data from memory to processing and back to memory; it takes a lot of time and energy to do that,” said Jianguo Mei, the Richard and Judith Wien Professor of Chemistry in Purdue University College of Science. “Our device has integrated functions of light perception, light to electric signal transformation, and on-site memory and data processing.” #TechTuesday feature: Traditional silicon architecture has propelled computer vision forward significantly, yet researchers at Purdue University are charting an alternative course, drawing inspiration from nature. Read more ➡ https://ow.ly/pIiE50QtgOS #Purdue #PurdueInnovates

After two years of work from Denis Lebold, Felix Schneider, and Fabian Schlenke under the Lead from Prof. Dr.-Ing. Michael Karagounis and Prof. Dr. Hendrik Wöhrle, we have finally released the OpenEye Hardware Accelerator for DNN Inference. It is based on the Eyeriss V2 Hardware Accelerator from the MIT. Features: * Implemented in pure Verilog * Row stationary dataflow * Computing with compressed weights and input activations * Flexible on-chip network * Scalable and flexible architecture that can be adapted using generics/parameters * Python-based front end for simple simulation https://lnkd.in/e-SBttFW

No longer than yesterday, Ted Neward gave a #keynote speech during GeeCON 2024. He argued that future of software engineering's paved by new languages that introduce revolutionary paradigms that will abstract away complexity of modern software development. Believe it or not, but it seems that we are witnesses of a new language being born. Higher Order Company introduced #Bend to abstract away complexities of parallel computing. Over two decades ago Sun introduced #JVM to abstract away many complexities of software development at that time like memory allocation and de-allocation. Now, Higher Order Company introduced #HVM to make running code on GPUs as simple as possible. The main idea behind it is to use a concurrent computation model devised in 1997 by Yves Lafont called Interaction Combinators. Ted in his speech specifically pointed out that academic research will drive this kind of an innovation. Ted, you're either a wizard or a very good expert as you predict the future of software engineering with remarkable accuracy.

How do you compare different #quantumcomputing solutions? 🤔 #benchmarking #QUARK #regulatoryaffairs #quantumtechnology has been one of the darlings of #research, with numerous challenges standing in the way of widescale adoption, not least the fact that the technical architecture itself requires incredibly specific environmental and physical conditions, not to mention #investment considerations. The ability to compare which technology and #quantum #algorithms offer practical benefits will increase in importance over time. #QUARK, or the quantum computing application benchmark, is an open-source framework for supporting the design, implementation, execution and analysis of benchmarks. The framework is written in #python and is made up of five components, of which the most interesting to me was the application component – offering the ability to integrate any #dataset (including #syntheticdata), then validate and evaluate provided solutions. Importantly, there is an assumption that classical systems will be used for validation, which is a view that may evolve over time, and will hopefully be reflected in the framework. I was a little surprised at the performance limitations of the various solutions in terms the number of nodes in the selected problems due to the quantum hardware, especially given that BMW was directly involved (I saw some of their impressive hardware and algorithms explained at #QuApps2022 – for #aerodynamics, if I'm not mistaken). If this is up your alley, I highly encourage you to sign up for #QuApps2024 organised by IVAM Microtechnology Network 👉 https://lnkd.in/ed5kFDrP Oh, and if you're wondering, the non-highlighted segments of the text are the stuff that I didn't have time to research, or didn't understand 😁 I will have the opportunity to connect with some #quantum researchers next week during the Infineon Winter School, but I have spent the last two weeks reading this paper in bits and pieces, and I don't want to miss the opportunity to share this fascinating research! Almost forgot to thank PD Dr. Jeanette Lorenz for giving me this delightful homework and to the research team, including Leonhard Hoelscher and Andre Luckow, for their incredible work and extremely well-written paper!

Announcing a new project in probabilistic computing, backed by the Advanced Research + Invention Agency (ARIA). Shannon Egan, Brock Doiron and Ashraf Lotfi proposed a radical new approach to probabilistic computing using the intrinsic properties of hardware itself. Very often, we ask computers to model probabilistic processes. This is a surprisingly inefficient process - asking machines designed for deterministic, binary operations to create random, more analogue outputs. In fact, much of computation's history has been spent trying to suppress the randomness of the physical world to avoid errors and 'noise'. The approach developed by Shannon, Brock and Ashraf flips a 'constraint' in computing, the rich, underlying randomness of physical devices, into an enabler of highly efficient probabilistic computing. The team, selected as one of 6 (so far) from 270 applicants across two rounds, will work with Suraj Bramhavar, a Programme Director at ARIA, as part of his opportunity space 'Nature Computes Better'. More in a link in comments.

Earlier this month I had the opportunity to attend the MIT Quantum Computing Hackathon (iQuHack) in a cohort of 210 hackers and placed 3rd for Quandela's challenge. I developed QForce with my partners Connor Totilas, Edward Zhang, Sameer Arora, and Shokhruz Kakharov. I want to thank Quandela for their support in the development process, the organizers of iQuHack, iQuise, and Massachusetts Institute of Technology for hosting. 💡 About our project: We developed a Quantum Generative Adversarial Network (QGAN) model coupled with a classical optimizer to demonstrate that qubits with a high fidelity can be achieved on Noisy Intermediate-Scale Quantum (NISQ) hardware using Quandela’s quantum photonic chip. Our results show that the model configuration is feasible even under conditions where up to 50% of the QGAN’s generator’s phase shifters are damaged and the discriminator’s phase shifters are subjected to phase noise up to 0.04π. 🔬 Key Highlights: 🌌 Circuit Optimization: QForce demonstrates the capabilities of advanced variationally trained quantum circuits, minimizing errors and maximizing qubit fidelity. 🤖 Generative Adversarial Learning: Through dynamic competition between generator and discriminator circuits, the model continuously enhances its performance. 📈 Classical Optimization: Our rigorous experimentation identified optimal classical optimizers to fine-tune quantum circuits and mitigate noise. 🔍 How we implemented QForce: ⚛️ Quantum Circuit: Drawing inspiration from Wang et al.'s pioneering work, we devised a variational circuit architecture tailored for QGAN implementation in photonics incorporating features such as phase changes and beam shifts. 🧠 Generator: Our generator circuit is designed to fabricate quantum Fock states, strategically aiming to replicate the true state and outmaneuver the discriminator. 🛠️ Discriminator: The discriminator evaluates input state measurements to discern between generated and true states. Its assessments provide insights into the effectiveness of the generator, driving iterative improvements in our QGAN model. 📈 Optimizer: After extensive experimentation, we opted to use Broyden-Fletcher-Goldfarb-Shanno (BFGS) and the limited memory variant (L-BFGS) due to its speed and relative precision. Learn more about QForce here:

Join Dan McCreary to explore how TigerGraph has integrated our massively parallel processing hashtag#graph and hashtag#AI platform with NVIDIA cuGraph to deliver 100x performance for accelerating graph algorithms. We'll cover: • architecture overview; • a multi-node, multi-GPU demonstration; • best practices

A new framework for simulating a distributed #quantum computing system, known as ARQUIN, was developed by 14 of #C2QA's partner institutions. The framework was split into "layers," so each research group could specialize in a specific aspect of the project. For example, some researchers focused on optimizing microwave-to-optical transduction while others designed algorithms that were compatible with the distributed architecture. The completed framework, in which all the layers come together, marks an important step toward efficient and scalable quantum computation. The team included researchers from Pacific Northwest National Laboratory, Brookhaven National Laboratory, the Massachusetts Institute of Technology, Yale University, Princeton University, Virginia Tech, IBM Quantum, and more. Learn about ARQUIN and the impressive research team: https://bit.ly/4dPaZ8t (Photo by Andrea Starr | Pacific Northwest National Laboratory)