Tensor Networks for Constrained Combinatorial Optimization

Item: Tensor Networks for Constrained Combinatorial Optimization
Rating: 4
Author: PostSingularity Institute

by Javier Lopez-Piqueres

Formal & Physical Sciences Mathematics PBMathematics PBUOptimization

🎙 Javier Lopez-Piqueres 👥 3K 📅 July 17, 2026 ⏱ 45 min 👁 324 📄 lecture 🧭 2026-08-15

Available in: English (current) Français

Keywords

tensor networkscombinatorial optimizationconstraintsMPSquantum-inspired

Summary

This WISER Summer Program session features Dr. Javier López-Piqueres, an applied research scientist at JPMorgan Chase, presenting his work on using tensor networks for constrained combinatorial optimization. He begins by motivating the importance of constraints in optimization, highlighting that constraints make problems hard. He reviews standard methods for handling constraints, such as penalty terms, relaxation, and sampling, noting their limitations. He then introduces tensor networks, focusing on matrix product states (MPS) and their ability to represent probability distributions. The core idea is to construct tensor networks that only generate feasible solutions by encoding constraints directly into the network structure. For equality constraints, he draws an analogy to U(1) symmetries in physics, where charge conservation leads to block-sparse structures. For inequality constraints, he introduces the concept of ‘quantum regions’ to handle bounds. He discusses optimization via a data-driven approach called the GO algorithm, which iteratively biases sampling towards low-cost regions. He presents results showing that symmetric tensor networks generalize better and are more efficient than vanilla MPS. He also contrasts tensor networks with neural networks, noting that internal symmetries are better handled by tensor networks due to multilinearity. The talk concludes with a summary and outlook, mentioning an open-source codebase for further exploration.

202 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides a clear and well-structured introduction to a novel approach for handling constraints in combinatorial optimization using tensor networks. The argumentation is solid, building from basic concepts to more advanced ideas. The speaker effectively motivates the need for constraint-aware models and demonstrates the advantages of tensor networks through examples and comparisons. The presentation is technical but accessible, with clear explanations of key concepts such as charge conservation and block-sparse structures. The use of the GO algorithm for optimization is well-explained, and the results show tangible benefits in terms of efficiency and generalization. The speaker also addresses potential limitations and contrasts with neural networks, strengthening the argumentation.

Scientific Rigor, Source Quality, Title Accuracy

The talk is scientifically rigorous, with the speaker referencing his own published works and related literature. He mentions specific papers and provides an open-source codebase for further study. The title accurately reflects the content, focusing on tensor networks for constrained combinatorial optimization. The presentation is well-organized and the technical details are consistent with established knowledge in tensor networks and quantum-inspired methods. However, as a lecture, it does not provide the same level of detail as a peer-reviewed paper, and some claims are not fully substantiated. The speaker also mentions a follow-up work, indicating ongoing research in the field.

221 words

Title / Content Match

The title accurately reflects the content: the talk focuses on tensor networks applied to constrained combinatorial optimization, covering both equality and inequality constraints.

Quality & Reliability

8/10

The talk is given by a researcher at JPMorgan Chase with a PhD in physics, presenting original research on tensor networks for constrained optimization. The methods are grounded in established physics concepts (symmetries, tensor networks) and the speaker provides references and an open-source codebase. However, the presentation is a lecture, not a peer-reviewed publication, and some claims are not independently verified.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

WISER — Mentioned as the program hosting the talk and for further information.

Concurring Sources

WISER — The talk is part of the WISER program, which aligns with the content.

Contribution & Novelties

The talk presents original research on using tensor networks to handle global constraints in combinatorial optimization, both equality and inequality. The key novelty is the construction of tensor networks that only generate feasible solutions by encoding constraints directly into the network structure, leveraging symmetries for equalities and introducing ‘quantum regions’ for inequalities. This approach contrasts with traditional methods that often rely on penalty terms or relaxation. The speaker also demonstrates the efficiency and generalization benefits of this approach through empirical results.

Pour aller plus loin :

Matrix product state — Background on MPS, a core concept in the talk.
Tensor network — General overview of tensor networks.
Combinatorial optimization — Context for the problem domain.
U(1) symmetry — The symmetry group relevant to equality constraints.
Quantum-inspired optimization — Related approaches.

130 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a technically strong and reliable presentation. The talk is well-balanced, with high information quality, technical depth, and reliability, though slightly lower on quantity of information due to the lecture format.

Reliability 8/10

💬 No comments were provided for analysis.