In this blog post we ask a simple question: can the Tiny Recursive Model (TRM), which solves difficult reasoning problems by reusing the same parameters recursively, also be useful for reinforcement learning? We adapted TRM into a Q-network and compared it with standard MLP and CNN architectures. The result was promising, but ultimately mixed: the recursive model sometimes learned faster early in training, but often converged to a worse solution than conventional architectures.

This work addresses two questions at the core of many battery applications: How can we reliably estimate the internal states of a battery cell from measurable quantities like voltage and current profiles, and how can we do that fast enough to be useful in real-world applications?

In this blog post we present jNO, a symbolic tracing layer built above JAX for neural-operator and PDE-oriented workflows. It unifies domains, model calls, residual expressions, and training objectives within a single traced program. Rather than treating geometry, model definition, and PDE constraints as separate stages, jNO expresses them in one symbolic layer and compiles them into one optimization pipeline.

Most physical phenomena — heat flow, wave propagation, fluid dynamics — are described mathematically by partial differential equations, or PDEs. The latest generation of AI models for physics takes inspiration from large language models. In our latest work, we show how to teach these so called foundation models the laws of physics.

This work addresses a great bottleneck in QAOA: How do we accurately estimate the performance of the algorithm based on the problem Hamiltonian before ever executing the circuit? In our recent work, we introduce QAOA predictor, that predicts the performance of LR-QAOA.

Last week at the 2026 Workshop on Reinforcement Learning (University of Mannheim), we presented a RL-based approach to a time-consuming problem in power electronics: the multi-objective optimization of power module topologies.

Accelerating the design of power transformers using convolutional physics-informed neural networks for fast, physics-compliant predictions of magnetic fields and core losses, developed by the AI-augmented Simulation research group at Fraunhofer IISB.

Announcing CleanQRL, an open-source library for lightweight Quantum Reinforcement Learning algorithms, developed by the AI-augmented Simulation research group at Fraunhofer IISB.

Exploring the advantages of PINNs in power electronics and semiconductor technology by the AI-augmented Simulation research group.