3 présentations

• 
  13h00 - 13h22

  Quantum-Inspired and Ising Architectures for Accelerated Optimization

  • Saavan Patel, prés., InfinityQ Technology

  This talk introduces InfinityQ’s TitanQ platform, a quantum-inspired approach for solving NP-Hard optimization problems. TitanQ leverages probabilistic Ising Machines, including techniques like Boltzmann Machines, to scale beyond 400,000 variables for mixed integer quadratic optimization. We will showcase its advantages through industrial and some academic applications. In finance, TitanQ achieves >1000x speedup compared to Gurobi in Mixed Integer Quadratically Constrained Quadratic Problems (MIQCQP) for index tracking. In logistics, it solves constrained clustering problems of >20,000 variables 100-1000x faster than Gurobi. Additionally, we present advancements in higher order binary optimization with TitanQ, achieving superior scaling (1.25^N) in the Low Autocorrelation Binary Sequences Problem, a notable academic challenge. These results highlight TitanQ’s industrial and academic breakthroughs.

• 
  13h22 - 13h44

  Optical Hardware for Quadratic Mixed Optimization Problems

  • Kirill Kalinin, prés., Microsoft Research
  • Hitesh Ballani, Microsoft Research
  • Christos Gkantsidis, Microsoft Research

  We develop an opto-electronic machine designed to accelerate optimization and machine learning applications. In optimization, our system addresses quadratic unconstrained mixed optimization (QUMO) problems by natively implementing continuous and binary variables in analog hardware. The QUMO abstraction enables efficient solutions to a wide range of realistic optimization challenges. This approach contrasts with many existing analog solvers, which focus on optimization problems limited to binary variables—such as the Ising model or quadratic unconstrained binary optimization (QUBO)—and often struggle with less efficient representations of practical applications. We evaluate our approach in a variety of practical problems including: (a) the transaction settlement problem, in collaboration with Barclays, a UK-based multinational bank; and (b) medical image reconstruction for MRI, in partnership with Microsoft Health Futures. In addition, we explore several neural network models that can be effectively implemented in analog hardware, as demonstrated by running the nonlinear regression and classification machine learning applications.

• 
  13h44 - 14h06

  Advancing Optimization with Entropy Quantum Computing (EQC)

  • Wesley Dyk, prés., Quantum Computing Inc.

  Entropy Quantum Computing (EQC), a novel paradigm developed by QCi, directly addresses optimization problems across various domains using an analog approach. Unlike traditional quantum hardware, which often requires complex mappings or auxiliary variables, EQC efficiently solves problems formulated as QUBO, Ising Hamiltonians, and higher-order polynomials with integer or continuous variables. This presentation will introduce the EQC paradigm, showcasing applications in portfolio optimization, flight trajectory planning, fraud detection, and feature selection. The talk will also explore the future potential of EQC, highlighting its ability to tackle mixed-integer problems and its miniaturization prospects through photonic chips made possible with thin-film lithium niobate.