Teaching

Master · 3 ECTS · 17151

Machine Learning for Quantum Systems

Prof. Dr. Annabelle Bohrdt

Thu 10:00–12:00 · Theresienstr. 37, A 348

Max. 15 participants · English · Registration by email

QST Master · 2 ECTS · 17323

QST-Fortgeschrittenenpraktikum: Representing (Fermionic) Quantum Many-Body States with Neural Networks

Prof. Dr. Annabelle Bohrdt

English · Registration by email with the lecturer

FoPra

Neural Network Quantum States

Coordinated by Fabian Döschl and Julius Tirpitz

This lab course introduces modern machine learning approaches to quantum many-body physics, focusing on the use of neural networks as variational ansätze for ground state wavefunctions. Starting from the exponential complexity of the many-body Schrödinger equation, the course motivates variational methods and reformulates ground state search as an optimization problem. Students implement Variational Monte Carlo (VMC) algorithms in which neural networks (Neural Quantum States) represent wavefunctions and are trained via stochastic sampling and gradient-based optimization.

In the first part, participants develop the VMC pipeline by studying the transverse-field Ising model, combining exact diagonalization benchmarks with neural-network-based approximations. Core techniques include Monte Carlo sampling via the Metropolis algorithm, evaluation of local observables, and optimization of variational energies.

The second part extends these methods to fermionic systems, addressing the challenges of antisymmetry through second quantization, Jordan–Wigner transformations, and Slater determinant constructions. Building on this, the course explores advanced architectures, such as neural-network-enhanced Slater determinants, to capture correlations in interacting systems like the t–V model.

Overall, the course provides a hands-on introduction to neural quantum states as a scalable framework for tackling strongly correlated quantum systems, bridging concepts from condensed matter physics and machine learning.

AI Lab

Quantum Reservoir Computing

Coordinated by Atiye Abedinnia

The idea of quantum computers dates back to Feynman's 1981 insight that only quantum systems can efficiently simulate other quantum systems, an observation that not only exposed the limitations of classical simulation but also sparked the foundational idea behind quantum computing. Since then, significant technological advances have pushed the field forward, enabling the development of various experimental platforms. Among the most prominent are trapped ions, superconducting qubits, and cold atom systems, including those based on Rydberg atoms, which are particularly promising due to their long coherence times and controllable interactions. Despite these advancements, the realization of large-scale, fault-tolerant quantum computers remains a significant challenge because of noise, decoherence, and scalability issues in current quantum hardware.

In parallel, the rise of artificial intelligence has motivated the development of Quantum Machine Learning (QML), which seeks to combine the rules of quantum computing with machine learning to efficiently learn and process complex data structures. However, many QML approaches, particularly variational quantum algorithms, are hindered by difficult optimization landscapes, including barren plateaus and gradient vanishing problems.

An alternative paradigm that has recently gained attention is Quantum Reservoir Computing (QRC). The idea in QRC is to replace the fixed, randomly connected hidden layer of a neural network with a dynamical quantum system. QRC relies on the inherent complexity of quantum dynamics. Classical data are injected into the quantum reservoir by encoding it into quantum states or by modulating Hamiltonian parameters. The system then evolves, generating rich, high-dimensional features due to quantum phenomena like superposition, entanglement, and interference. Instead of training the internal dynamics, only the linear readout layer is trained, typically through linear regression based on measurements of observables from the quantum system. This gradient-free approach enables QRC to perform various machine learning tasks, while remaining well suited for near-term quantum hardware.

Kolloquium · 17152

Bohrdt Group Seminar

Prof. Dr. Annabelle Bohrdt

Wed 12:00–14:00 · Theresienstr. 37, A 348

Kolloquium · 17322

Journal Club: Machine Learning for Condensed Matter Physics

Prof. Dr. Annabelle Bohrdt

Tue 10:00–12:00 · selected dates (see LSF for schedule)