Wolfgang Gentzsch

"A new study conducted at the University of Vienna, the Max Planck Institute for Intelligent Systems in Stuttgart, and the Helmholtz Centers in Berlin and Dresden takes an important step in the challenge to miniaturize computing devices and to make them more energy-efficient. The work, published in Science Advances, opens up new possibilities for creating reprogrammable magnonic circuits by exciting spin waves by alternating currents and redirecting these waves on demand. The central processing units (CPUs) that we use in our laptops, desktops or even phones use billions of transistors, which are based on the complementary metal oxide semiconductor (CMOS) technology. With the increasing need to miniaturize these devices, several physical limitations are raising concerns about their sustainability. In addition, high power consumption and energy losses push scientists to search for alternative computing architectures." #magnonic #spinwaves

2

Day 16 of #geekstreak2024 Problem - K Sized Subarray Maximum Continuing my 21-day POTD Challenge, "#geekstreak2024" powered by Deutsche Bank! Today’s challenge is "K Sized Subarray Maximum." Problem: Given an array arr[] and an integer k, the task is to find the maximum for each and every contiguous subarray of size k. Example 1: Input: k = 3, arr[] = [1, 2, 3, 1, 4, 5, 2, 3, 6] Output: [3, 3, 4, 5, 5, 5, 6] Explanation: 1st subarray = [1, 2, 3] → max = 3 2nd subarray = [2, 3, 1] → max = 3 3rd subarray = [3, 1, 4] → max = 4 And so on... Example 2: Input: k = 4, arr[] = [8, 5, 10, 7, 9, 4, 15, 12, 90, 13] Output: [10, 10, 10, 15, 15, 90, 90] Explanation: 1st subarray = [8, 5, 10, 7] → max = 10 2nd subarray = [5, 10, 7, 9] → max = 10 3rd subarray = [10, 7, 9, 4] → max = 10 And so on... https://lnkd.in/diKySy26 Approach: To solve this problem efficiently in O(n) time: I implemented the sliding window technique using a Deque. For each element, the deque maintains the indices of the elements in such a way that the maximum for the current window is always at the front of the deque. The deque only stores useful elements for the current window, and elements that go out of the window are removed. This approach efficiently calculates the maximum in each subarray of size k in O(n) time and uses O(k) space. Tags: #geekstreak2024 #DeutscheBank #CodingChallenge #ProblemSolving #POTD #GeeksForGeeks #JavaCoding #SlidingWindow #ArrayProblem #InterviewPreparation GeeksforGeeks Deutsche Bank Consistency is key! For the next 5 days, I’ll continue solving the Problem of the Day (POTD) on GeeksforGeeks to further enhance my problem-solving skills. Stay tuned for more updates on this journey! 🚀 Find all my daily code here: https://lnkd.in/dRxPhVJa

6

"Energy-based Hopfield Boosting for Out-of-Distribution Detection" by Claus Hofmann, Simon Schmid, Sepp Hochreiter et al. "Out-of-distribution (OOD) detection is critical when deploying machine learning models in the real world. Outlier exposure methods, which incorporate auxiliary outlier data in the training process, can drastically improve OOD detection performance compared to approaches without advanced training strategies. We introduce Hopfield Boosting, a boosting approach, which leverages modern Hopfield energy (MHE) to sharpen the decision boundary between the in-distribution and OOD data. Hopfield Boosting encourages the model to concentrate on hard-to-distinguish auxiliary outlier examples that lie close to the decision boundary between in-distribution and auxiliary outlier data. Our method achieves a new state-of-the-art in OOD detection with outlier exposure, improving the FPR95 metric from 2.28 to 0.92 on CIFAR-10 and from 11.76 to 7.94 on CIFAR-100." #machinelearning

31

"Researchers from Skoltech, Jilin University and Beijing HPSTAR in China, and their German colleagues have synthesized and studied a new type of hydrogen-rich superconductor. Technically referred to as an A15-type lanthanum superhydride, with the formula La4H23, it shows superconductivity below minus 168 degrees Celsius at a pressure of 1.2 million atmospheres. The research results were published in the National Science Review. Polyhydrides are a novel class of compounds synthesized at about 1 million times the normal atmospheric pressure on Earth. They can exhibit unique superconducting properties with record-high critical temperatures of up to -23 C in lanthanum decahydride LaH10, critical magnetic fields reaching 300 tesla, and critical current densities." #materialscience #superconductors

2

In the article "Scaling ChatGPT: Five Real-World Engineering Challenges", Gergely Orosz delves into the complexities of making ChatGPT a scalable and reliable tool. One of the primary challenges discussed is the significant computational demand of running large language models (LLMs). This involves balancing the need for high-performance hardware with the practical constraints of cost and energy efficiency. Orosz explains how the engineering team at OpenAI optimizes the model's inference stage to ensure quick response times while maintaining accuracy and coherence in the generated text​. https://lnkd.in/gH-Z6Vx7

4

💥💥💥 Free Energy Projective Simulation (FEPS): Active inference with interpretability Joséphine Pazem, Marius Krumm, Alexander Q. Vining, Lukas J. Fiderer, Hans J. Briegel Abstract In the last decade, the free energy principle (FEP) and active inference (AIF) have achieved many successes connecting conceptual models of learning and cognition to mathematical models of perception and action. This effort is driven by a multidisciplinary interest in understanding aspects of self-organizing complex adaptive systems, including elements of agency. Various reinforcement learning (RL) models performing active inference have been proposed and trained on standard RL tasks using deep neural networks. Recent work has focused on improving such agents' performance in complex environments by incorporating the latest machine learning techniques. In this paper, we take an alternative approach. Within the constraints imposed by the FEP and AIF, we attempt to model agents in an interpretable way without deep neural networks by introducing Free Energy Projective Simulation (FEPS). Using internal rewards only, FEPS agents build a representation of their partially observable environments with which they interact. Following AIF, the policy to achieve a given task is derived from this world model by minimizing the expected free energy. Leveraging the interpretability of the model, techniques are introduced to deal with long-term goals and reduce prediction errors caused by erroneous hidden state estimation. We test the FEPS model on two RL environments inspired from behavioral biology: a timed response task and a navigation task in a partially observable grid. Our results show that FEPS agents fully resolve the ambiguity of both environments by appropriately contextualizing their observations based on prediction accuracy only. In addition, they infer optimal policies flexibly for any target observation in the environment. 👉 https://lnkd.in/dJxV7QXK #machinelearning

3

🗣 Had a great time on a high energy panel at the Supercomputing Conference 2024, discussing "Educating for a Hybrid Future: Bridging the Gap between High-Performance and Quantum Computing" 🌉 Honoured to share the stage with Dr. Stefan Seegerer (IQM Quantum Computers), Monica VanDieren (NVIDIA), Bruno Abreu (Pittsburgh Supercomputing Center), and Minh Chung (Leibniz Supercomputing Centre ), and explore the exciting intersection of quantum computing 🖥 and HPC. From discussing the role of humanities students in quantum science 🔭 to the importance of diversity 👦👩 and inclusion 👦👦 in the field, we covered a remarkable spectrum of topics in the short 90 minutes 🕐 We also shared insights on various educational initiatives, workforce development programs, and the fundamental role of math and physics ⚖️ in building a strong foundation for quantum computing and HPC. I was a bundle of nerves before we took the stage. Thanks to the artful moderation by Vincent A. 👏 and the lively dialogue with the audience, I quickly eased into the moment and found myself truly enjoying the session. It was a reminder that stepping outside of your comfort zone can lead to incredible experiences 👣 Looking forward to the next Supercomputing Conference in St Louis, 2025! 🚀 National Research Foundation Singapore National Supercomputing Centre (NSCC) Singapore Institute of High Performance Computing (A*STAR IHPC) National Quantum Computing Hub (NQCH) #marvellouslyyours #sc24 #hpc #quantumtech #quantum #quantumcomputing #tech #research #outreach #education #math #physics #humanities #training #development #nvidia #aws #iqm #quantinuum #lsc #psc #nrf #nscc #workforce #ai

79

3 Kommentare

The Entropy of Time: The Clock Conundrum Limiting Quantum Computing’s Future TOPICS:Quantum ComputingQuantum Information ScienceVienna University Of Technology By VIENNA UNIVERSITY OF TECHNOLOGY JANUARY 7, 2024 Quantum computers require precise time measurement for manipulating quantum states, but new research has revealed a fundamental limitation: clocks cannot achieve perfect resolution and precision simultaneously due to their finite energy and entropy generation. This imposes inherent constraints on quantum computing’s capabilities. As quantum technology advances, overcoming these time measurement challenges will become critical, possibly leading to new discoveries in quantum mechanics.

1

4 Kommentare

Intersect360 Research has released a comprehensive new report concerning the challenges facing the combined fields of high-performance computing (HPC) and artificial intelligence (AI). Titled “Issues Facing the HPC-AI Industry: Insights from the Advisory Committees of the HPC-AI Leadership Organization (HALO),” this report – which was based on extensive interviews conducted with members of the HALO Advisory Committee – details the industry’s current landscape and future directions of HPC and AI technologies. Full report: https://lnkd.in/eJeZ--e3 Webinar: https://lnkd.in/em5TSh5c While the full report is 22 pages long and complete with figures and graphs, certain information and opinions stand out. To begin, there is currently a lack of people who are trained and ready to handle new HPC-AI environments. Intersect360 Research interviews found that most respondents expressed concerns about a shortage of personnel, particularly in the computational sciences. This problem stems from and is exacerbated by educational programs that do not emphasize computational sciences. Both computational/engineering sciences as well as the design and operation of large-scale computational infrastructures suffer from this shortage of knowledgeable professionals. Additionally, interviewees remarked that while GPU-heavy deployments often used in AI are capable of delivering a large number of theoretical peak FLOPS, many traditional HPC applications do not scale well on large numbers of GPUs. This finding is consistent with broader Intersect360 Research studies, which found that 12% of HPC-AI users report “moderate” or “major” performance issues with their HPC-AI systems, while only 47% report “no problems.” Accuracy and the reproducibility of results were also both major points of contention within the report. The variety of chip technologies – including CPUs, GPUs, NPUs, FPGAs, and custom chips – is introducing complexity in HPC-AI environments that can lead to an inability to achieve consistent results and performance across different architectures. Diverse chip technologies and software stacks also create issues with independent verification of research, thereby requiring a large investment of resources to replicate results across different HPC-AI environments.

2

Creator of ARC challenge. This interview with François Chollet is fascinating. He talked about transduction and induction approaches (and relationship between the 2) along with program graph structures being created with advanced searches working alongside gradients to get to more likely answers. This is an orchestration. Extremely complex and complicated in the front view mirror and will become just easy as we move past the goal of achievable. **He gets to this point where he states that you will eventually not need to search in a way.** I sat with that a moment and he is talking about looking at things “whole” and I to believe there is a way to understand without search as we see it today but I’m unsure it’s possible. But isn’t looking for the impossible such a fun endeavor! My words are inaccurate as the anthropomorphizing is incredible due to my language constraints. #ai #llm #genai #graphs #induction #transduction Video: https://lnkd.in/gtkJ2umc Kelly Cohen Lohith Srikanth Pentapalli Scott Phillips Eduardo Ordax Lori Mazor Gopal Padinjaruveetil Joe Oquist💡 Zachary Huhn Summer Crenshaw Josette Riep Machine Learning Street Talk (MLST) - Great interview again. Always looking up something in the conversation and learning something new. Pushing boundaries.

22

2 Kommentare

A new proof has brought mathematicians one step closer to understanding the hidden order of those “atoms of arithmetic,” the prime numbers. https://lnkd.in/gtEn-XhB

2

It's not sufficient to avoid downtime and keep production losses minimal in this era. The metric is not just production uptime anymore. Now, it's about keep improving: Production/kWH of Energy consumed & Production/tonne of CO2 emitted. #netzero is not a buzzword anymore, proving with metrics like above that one is moving in the direction towards Net-Zero is key to avail loans at favourable interest rates from top 10 money managers like BlackRock, VanGuard, that basically run the world. (https://lnkd.in/gJ_Ed392) With our #energycenteredmaintenance approach, we are helping you to keep your Production/kWH of energy consumed and Production/tonne of CO2 emitted improving. If you are attending the Amazon Web Services (AWS) stall at HANNOVER MESSE, consider taking a detour to Hall 7 - Booth D28, Nanoprecise Sci Corp, an AWS partner solution for Energy Centric Predictive Maintenance. We will talk to you about an investment policy that atleast 2X's your money in a year while helping you trend towards NetZero. Sounds too good to be true? Lets talk, we are putting our money where our mouth is.  https://lnkd.in/gv5y8iW2 Qais Aldababneh, Kalyan Meduri #reliability #maintenance #motors #ghg #efficiency #conditionmonitoring

64

1 Kommentar

What makes high temperature superconductivity possible? Researchers get closer to a unified theory https://lnkd.in/ep3fxY6g

1