IPAB Workshop - 16/4/26 Speaker: Zhaocheng Liu Title: A Geometry-Based Approach for Computing Protein Transition Pathways from Molecular Dynamics SamplesAbstract: Protein transition pathways are critical for understanding protein folding and conformational changes. That is, given two conformations of a protein, what are the most likely intermediate conformations between them. However, finding these pathways is computationally challenging due to the intractable complexity in both protein dynamics and dimensionality. In this work, we propose a new geometry-based paradigm with an algorithm to efficiently compute pathways from molecular dynamics samples. We demonstrate we can discover pathways for a 370-amino-acids protein, whereas existing works up to 58 amino acids. Our method works at full atoms precision instead of in a latent space, and shows promising capability to scale to even larger proteins. Speaker: Qiyue Xia Title: Interpretability by Design for Efficient Multi-Objective Reinforcement Learning Abstract: Multi-objective reinforcement learning (MORL) aims at optimising several, often conflicting goals to improve the flexibility and reliability of RL in practical tasks. This is typically achieved by finding a set of diverse, non-dominated policies that form a Pareto front in the performance space. We introduce LLE-MORL, an approach that achieves interpretability by design by utilising a training scheme based on the local relationship between the parameter space and the performance space. By exploiting a locally linear map between these spaces, our method provides an interpretation of policy parameters in terms of the objectives, and this structured representation enables an efficient search within contiguous solution domains, allowing for the rapid generation of high-quality solutions without extensive retraining. Experiments across diverse continuous control domains demonstrate that LLE-MORL consistently achieves higher Pareto front quality and efficiency than state-of-the-art approaches. Speaker: Heba Al Kayed Title: Compositional Verification of Learning-Enabled Autonomous Systems Abstract: Verifying the safety of learning-enabled autonomous systems remains a critical challenge due to the stochastic environments they operate in and the uncertainty inherent in the Deep Neural Networks (DNNs) used for perception . In this talk, I will discuss the common challenges in verifying such systems, particularly the complexities introduced by hybrid dynamics and component heterogeneity. I will then present our ongoing work on a compositional verification framework designed to address these issues. I will demonstrate our current progress for an Autonomous Emergency Braking (AEB) system, illustrating how we aim to guarantee safety. Apr 16 2026 13.00 - 14.00 IPAB Workshop - 16/4/26 Zhaocheng Liu, Qiyue Xia & Heba Al Kayed MF2
IPAB Workshop - 16/4/26 Speaker: Zhaocheng Liu Title: A Geometry-Based Approach for Computing Protein Transition Pathways from Molecular Dynamics SamplesAbstract: Protein transition pathways are critical for understanding protein folding and conformational changes. That is, given two conformations of a protein, what are the most likely intermediate conformations between them. However, finding these pathways is computationally challenging due to the intractable complexity in both protein dynamics and dimensionality. In this work, we propose a new geometry-based paradigm with an algorithm to efficiently compute pathways from molecular dynamics samples. We demonstrate we can discover pathways for a 370-amino-acids protein, whereas existing works up to 58 amino acids. Our method works at full atoms precision instead of in a latent space, and shows promising capability to scale to even larger proteins. Speaker: Qiyue Xia Title: Interpretability by Design for Efficient Multi-Objective Reinforcement Learning Abstract: Multi-objective reinforcement learning (MORL) aims at optimising several, often conflicting goals to improve the flexibility and reliability of RL in practical tasks. This is typically achieved by finding a set of diverse, non-dominated policies that form a Pareto front in the performance space. We introduce LLE-MORL, an approach that achieves interpretability by design by utilising a training scheme based on the local relationship between the parameter space and the performance space. By exploiting a locally linear map between these spaces, our method provides an interpretation of policy parameters in terms of the objectives, and this structured representation enables an efficient search within contiguous solution domains, allowing for the rapid generation of high-quality solutions without extensive retraining. Experiments across diverse continuous control domains demonstrate that LLE-MORL consistently achieves higher Pareto front quality and efficiency than state-of-the-art approaches. Speaker: Heba Al Kayed Title: Compositional Verification of Learning-Enabled Autonomous Systems Abstract: Verifying the safety of learning-enabled autonomous systems remains a critical challenge due to the stochastic environments they operate in and the uncertainty inherent in the Deep Neural Networks (DNNs) used for perception . In this talk, I will discuss the common challenges in verifying such systems, particularly the complexities introduced by hybrid dynamics and component heterogeneity. I will then present our ongoing work on a compositional verification framework designed to address these issues. I will demonstrate our current progress for an Autonomous Emergency Braking (AEB) system, illustrating how we aim to guarantee safety. Apr 16 2026 13.00 - 14.00 IPAB Workshop - 16/4/26 Zhaocheng Liu, Qiyue Xia & Heba Al Kayed MF2
