Applying Quantum-Inspired Techniques in Engineering

🔬 What if the very surface of a device could think, sense, and communicate — all at once? After 3 years of exploration at the intersection of computing and communication, I’m thrilled to share a summary of our key findings in my recent talk: “The New Frontier in Neural Network Surfaces" In this work, we dive deep into how Rydberg atom arrays and stacked intelligent metasurfaces (SIMs) are redefining the boundary between physics and machine learning. Instead of placing AI on top of hardware — we’re embedding intelligence into the physical layer itself. Imagine surfaces that: Adapt instantly to their environment Perform beamforming and sensing without external computation Learn and compute directly through engineered quantum and wave interactions We explore how Rydberg atom arrays can be harnessed to create neural network surfaces — programmable, adaptive platforms that unify sensing, computation, and communication at the physical layer. 🌍 The implications are vast: from 6G wireless networks, to edge AI, to quantum-inspired computing that operates with ultra-low energy and unprecedented speed. I’m proud of the brilliant collaborators who’ve made this journey possible and grateful for the recognition it’s already received — including a Best Paper Award at ICC 2023. #AI #NeuralSurfaces #6G #QuantumComputing #EdgeAI #Metasurfaces #RydbergAtoms #FutureNetworks #WaveDomainAI #RAQR #SIM

🌟 Exciting News! 🌟 I'm thrilled to announce that our latest research on tackling the challenging Low Autocorrelation Binary Sequences (LABS) problem has been published! Check it out here: https://lnkd.in/e2Q3qaEc. Our scientific article, titled "New Improvements in Solving Large LABS Instances Using Massively Parallelizable Memetic Tabu Search," showcases groundbreaking advancements. LABS is a complex binary optimization problem with significant applications in telecommunications, physics, math and finance. Solving it becomes particularly difficult for sizes beyond 66. In our work, we introduce a massively parallelized GPU implementation of the memetic tabu search algorithm, capable of handling LABS sizes up to 120. By leveraging block-level and thread-level parallelism on a single NVIDIA A100 GPU and developing hyper-optimized binary-valued data structures for shared memory, we achieved up to a 26-fold speedup over a 16-core CPU implementation. Our implementation led to the discovery of new LABS energy values for twelve problem sizes between 92 and 118. Notably, we improved values for two odd-sized problems (99, 107), challenging the previous best-known results obtained using the skew-symmetry property. This highlights that relying solely on skew-symmetry may lead to suboptimal solutions in the quest for global optima. Our findings underscore the potential of accelerating powerful meta-heuristics to find near-optimal solutions for combinatorial optimization problems within a fixed time budget. This result comes from the #QuantumInspired Algorithms Group, led by Niraj Kumar, in Global Technology Applied Research at JPMorganChase. Quantum-inspired algorithms are classical algorithms that are designed to mimic certain aspects of #QuantumComputing to solve complex problems more efficiently than traditional classical algorithms. Being classical, these algorithms do not require #quantum hardware; instead, they leverage principles and techniques inspired by quantum mechanics to achieve performance improvements on classical computers. Therefore, unlike quantum algorithms, quantum-inspired algorithms can be used in production today! Stay tuned for more quantum-inspired results in the near future! Coauthors: Zhiwei Zhang, Jiayu Shen, Niraj Kumar, and Marco Pistoia.

Excited to share the latest breakthrough in quantum-inspired computing! Our team has just published a groundbreaking paper in the Journal of Cheminformatics titled "Application of the digital annealer unit in optimizing chemical reaction conditions for enhanced production yields." Read the full paper here: https://lnkd.in/gdzYBDhb In the fast-paced world of chemical synthesis and pharmaceutical development, finding optimal reaction conditions to maximize yields is a massive challenge due to the enormous chemical space involved. Traditional experiments are time-consuming and resource-intensive, while even advanced machine learning (ML) models struggle with the computational demands of exploring all possible combinations. This paper introduces an innovative hybrid approach: the Digital Annealer Unit (DAU)—a quantum-inspired hardware accelerator—to solve Quadratic Unconstrained Binary Optimization (QUBO) problems for predicting and optimizing reaction yields. By constructing two types of QUBO models (one based on quantum annealing principles and another integrated with ML), this approach achieves performance on par with classical ML methods like Random Forest and Multilayer Perceptron, but with inference times reduced to mere seconds. Key findings include: Comparable Accuracy with Superior Speed: Our models were tested on high-throughput experimentation (HTE) datasets and Reaxys data, showing robust yield predictions while dramatically accelerating combinatorial optimization—screening billions of conditions millions of times faster than conventional CPUs. Active Learning Integration: In simulated campaigns, strategically selecting data points via active learning led to faster convergence on high-yield conditions, with performance improvements plateauing after just a few iterations. Real-World Implications: This method addresses data scarcity and complexity in reactions (e.g., C-N cross-coupling), paving the way for autonomous design in labs and faster innovation in drug discovery. This study demonstrates the application of DAUs to efciently optimize chemical reaction conditions, leverag‑ ing quadratic unconstrained binary optimization (QUBO) models for accurate yield predictions. The QUBO-based approach exhibits comparable performance to classical machine learning methods while achieving inference times in seconds, significantly accelerating the screening of billions of reaction conditions. By integrating active learning and DAU technology, this research establishes a novel framework for reaction condition optimization, enabling innovative advancements in chemical synthesis. A huge spotlight on Jimmy Yen-Chu Lin , whose visionary leadership was central to this project.

Bridging Physics and AI: A Quantum-Inspired Leap in Neural Network Optimization I've mapped a transformer model (SmolLM2-135M) to an Ising spin system and achieved unprecedented optimization using Q*Agents - Heated Ballistic Bifurcation (Quantum Tunneling). What’s New? 🧠 Reimagining AI Optimization: I treated the transformer’s weights as a digital twin of a physical system, governed by quantum principles. Leveraged Q*agents in a quantum-inspired algorithm to reveal emergent structural patterns in neural networks. 📊 Results That Speak for Themselves: Chaotic weight distributions transformed into organized, periodic patterns. Quantum tunneling-like effects drove efficient optimization, even in high-dimensional spaces. Evidence of emergent organization—is this the physics of intelligence manifesting in AI? Why It Matters: 🔋 From Theory to Reality: Imagine training massive AI models with quantum efficiency—on classical hardware. These results pave the way for next-gen optimization techniques that blend physics and machine learning. ⚡ Key Technical Highlights: Model: SmolLM2-135M transformed via Ising spin mapping Method: Heated Ballistic Bifurcation Innovation: Emergent banding in weight matrices suggests fundamental principles shaping model architectures 🌌 This is only the beginning. Are physics-inspired approaches the next frontier in AI? Could this redefine how we design, train, and optimize large models? Let’s collaborate, discuss, and explore these exciting intersections. Share your insights or connect if you’re working in this space!