Physics-Informed Neural Networks.

This paper extends biologically-informed neural networks (BINNs) from 1D to 2D spatial domains for learning reaction-diffusion equations from data, combining neural network training with symbolic regression to discover closed-form equations. The method is demonstrated on real lung cancer cell microscopy data, successfully recovering interpretable 2D+time reaction-diffusion models.

This paper uses Deep Operator Networks (DeepONets) to learn a surrogate for the Riccati differential equation solution operator in finite-horizon LQR problems, enabling fast approximate optimal control without repeated numerical integration. The approach includes theoretical guarantees on stability and performance, and demonstrates significant computational speedups while maintaining high accuracy.

This paper introduces Adaptive Conformal Filtering (ACoFi), which combines learned safety filters with adaptive conformal inference to provide soft safety guarantees for control systems. The method dynamically adjusts switching criteria between nominal and safe policies based on prediction uncertainty, achieving better safety performance than fixed-threshold approaches.

This paper provides a complete, step-by-step manual derivation of how Physics-Informed Neural Networks (PINNs) are trained, including forward propagation, loss computation, and backpropagation with explicit numerical examples. It bridges the gap between automatic differentiation libraries and the underlying mathematical operations, making PINN training transparent and verifiable.

This paper proposes CmIR, a causal inference framework that separates multimodal data into stable causal features and spurious environment-specific features to improve robustness in affective computing. The method achieves state-of-the-art performance, especially on out-of-distribution and noisy data.

This paper develops a random matrix theory model that explains why neural networks exhibit a transient learning window where signal is detectable before overfitting occurs. The key mechanism is that anisotropy in input data creates fast and slow learning directions, causing a learnable eigenvalue to temporarily separate from noise before being reabsorbed.

This paper develops a hybrid simulation framework for large HVAC systems that combines physics-informed neural networks with traditional differential-algebraic equation solvers, achieving multi-fold speedups over high-fidelity simulations while maintaining low errors. The approach scales to systems with 32+ compressor-condenser pairs by learning component-level dynamics and enforcing system-level physical constraints.

This paper shows that randomly initialized neural networks can learn useful representations through simple peer-to-peer consensus (self-distillation) alone, without projectors, predictors, or pretext tasks. The findings suggest that self-distillation itself is a key mechanism driving learning in self-supervised methods, independent of other architectural components.

This paper presents a self-supervised deep learning method using bilateral wrist-worn IMU sensors to detect Parkinson's disease, achieving over 93% accuracy for distinguishing PD from healthy controls and demonstrating effective transfer learning with only 20% labeled data. The model is lightweight enough to run in real-time on a Raspberry Pi, making it practical for clinical deployment.

DiLaR-PINN is a physics-informed neural network that learns unmodeled dissipative effects in electromechanical systems by constraining residual terms to be energy-dissipating rather than energy-injecting. The method combines first-principles models with data-driven components that operate on latent states and guarantee physically consistent energy behavior.

This paper applies Hebbian learning principles to audio classification in a continual learning setting, introducing a kernel plasticity approach that selectively updates network weights to balance learning new sounds while retaining knowledge of previously learned ones. On the ESC-50 dataset, the method achieves 76.3% accuracy across five incremental learning steps, significantly outperforming a baseline approach.