Since its foundations, more than one hundred years ago, the field of structural biology has strived to understand and analyze the properties of molecules and their interactions by studying the structure that they take in 3D space. However, a fundamental challenge with this approach has been the dynamic nature of these particles, which forces us to model not a single but a whole distribution of structures for every molecular system. This thesis proposes Intrinsic Diffusion Modeling, a novel approach to this problem based on combining diffusion generative models with scientific knowledge about the flexibility of biological complexes. The knowledge of these degrees of freedom is translated into the definition of a manifold over which the diffusion process is defined. This manifold significantly reduces the dimensionality and increases the smoothness of the generation space allowing for significantly faster and more accurate generative processes. We demonstrate the effectiveness of this approach on two fundamental tasks at the basis of computational chemistry and biology: molecular conformer generation and molecular docking. In both tasks, we construct the first deep learning method to outperform traditional computational approaches achieving an unprecedented level of accuracy for scalable programs.

Solving high-dimensional Bayesian inverse problems (BIPs) with the variational inference (VI) method is promising but still challenging. The main difficulties arise from two aspects. First, VI methods approximate the posterior distribution using a simple and analytic variational distribution, which makes it difficult to estimate complex spatially-varying parameters in practice. Second, VI methods typically rely on gradient-based optimization, which can be computationally expensive or intractable when applied to BIPs involving partial differential equations (PDEs). To address these challenges, we propose a novel approximation method for estimating the high-dimensional posterior distribution. This approach leverages a deep generative model to learn a prior model capable of generating spatially-varying parameters. This enables posterior approximation over the latent variable instead of the complex parameters, thus improving estimation accuracy. Moreover, to accelerate gradient computation, we employ a differentiable physics-constrained surrogate model to replace the adjoint method. The proposed method can be fully implemented in an automatic differentiation manner. Numerical examples demonstrate two types of log-permeability estimation for flow in heterogeneous media. The results show the validity, accuracy, and high efficiency of the proposed method.

Current embedded systems are specifically designed to run multimedia applications. These applications have a big impact on both performance and energy consumption. Both metrics can be optimized selecting the best cache configuration for a target set of applications. Multi-objective optimization may help to minimize both conflicting metrics in an independent manner. In this work, we propose an optimization method that based on Multi-Objective Evolutionary Algorithms, is able to find the best cache configuration for a given set of applications. To evaluate the goodness of candidate solutions, the execution of the optimization algorithm is combined with a static profiling methodology using several well-known simulation tools. Results show that our optimization framework is able to obtain an optimized cache for Mediabench applications. Compared to a baseline cache memory, our design method reaches an average improvement of 64.43\% and 91.69\% in execution time and energy consumption, respectively.

We present algorithms for uniformly covering an unknown indoor region with a swarm of simple, anonymous and autonomous mobile agents. The exploration of such regions is made difficult by the lack of a common global reference frame, severe degradation of radio-frequency communication, and numerous ground obstacles. We propose addressing these challenges by using airborne agents, such as Micro Air Vehicles, in dual capacity, both as mobile explorers and (once they land) as beacons that help other agents navigate the region. The algorithms we propose are designed for a swarm of simple, identical, ant-like agents with local sensing capabilities. The agents enter the region, which is discretized as a graph, over time from one or more entry points and are tasked with occupying all of its vertices. Unlike many works in this area, we consider the requirement of informing an outside operator with limited information that the coverage mission is complete. Even with this additional requirement we show, both through simulations and mathematical proofs, that the dual role concept results in linear-time termination, while also besting many well-known algorithms in the literature in terms of energy use.

Transfer learning has recently become the dominant paradigm of machine learning. Pre-trained models fine-tuned for downstream tasks achieve better performance with fewer labelled examples. Nonetheless, it remains unclear how to develop models that specialise towards multiple tasks without incurring negative interference and that generalise systematically to non-identically distributed tasks. Modular deep learning has emerged as a promising solution to these challenges. In this framework, units of computation are often implemented as autonomous parameter-efficient modules. Information is conditionally routed to a subset of modules and subsequently aggregated. These properties enable positive transfer and systematic generalisation by separating computation from routing and updating modules locally. We offer a survey of modular architectures, providing a unified view over several threads of research that evolved independently in the scientific literature. Moreover, we explore various additional purposes of modularity, including scaling language models, causal inference, programme induction, and planning in reinforcement learning. Finally, we report various concrete applications where modularity has been successfully deployed such as cross-lingual and cross-modal knowledge transfer. Related talks and projects to this survey, are available at https://www.modulardeeplearning.com/.

Despite their growing popularity, data-driven models of real-world dynamical systems require lots of data. However, due to sensing limitations as well as privacy concerns, this data is not always available, especially in domains such as energy. Pre-trained models using data gathered in similar contexts have shown enormous potential in addressing these concerns: they can improve predictive accuracy at a much lower observational data expense. Theoretically, due to the risk posed by negative transfer, this improvement is however neither uniform for all agents nor is it guaranteed. In this paper, using data from several distributed energy resources, we investigate and report preliminary findings on several key questions in this regard. First, we evaluate the improvement in predictive accuracy due to pre-trained models, both with and without fine-tuning. Subsequently, we consider the question of fairness: do pre-trained models create equal improvements for heterogeneous agents, and how does this translate to downstream utility? Answering these questions can help enable improvements in the creation, fine-tuning, and adoption of such pre-trained models.

We introduce RoboNinja, a learning-based cutting system for multi-material objects (i.e., soft objects with rigid cores such as avocados or mangos). In contrast to prior works using open-loop cutting actions to cut through single-material objects (e.g., slicing a cucumber), RoboNinja aims to remove the soft part of an object while preserving the rigid core, thereby maximizing the yield. To achieve this, our system closes the perception-action loop by utilizing an interactive state estimator and an adaptive cutting policy. The system first employs sparse collision information to iteratively estimate the position and geometry of an object's core and then generates closed-loop cutting actions based on the estimated state and a tolerance value. The "adaptiveness" of the policy is achieved through the tolerance value, which modulates the policy's conservativeness when encountering collisions, maintaining an adaptive safety distance from the estimated core. Learning such cutting skills directly on a real-world robot is challenging. Yet, existing simulators are limited in simulating multi-material objects or computing the energy consumption during the cutting process. To address this issue, we develop a differentiable cutting simulator that supports multi-material coupling and allows for the generation of optimized trajectories as demonstrations for policy learning. Furthermore, by using a low-cost force sensor to capture collision feedback, we were able to successfully deploy the learned model in real-world scenarios, including objects with diverse core geometries and soft materials.

Artificial intelligence (AI)-empowered industrial fault diagnostics is important in ensuring the safe operation of industrial applications. Since complex industrial systems often involve multiple industrial plants (possibly belonging to different companies or subsidiaries) with sensitive data collected and stored in a distributed manner, collaborative fault diagnostic model training often needs to leverage federated learning (FL). As the scale of the industrial fault diagnostic models are often large and communication channels in such systems are often not exclusively used for FL model training, existing deployed FL model training frameworks cannot train such models efficiently across multiple institutions. In this paper, we report our experience developing and deploying the Federated Opportunistic Block Dropout (FEDOBD) approach for industrial fault diagnostic model training. By decomposing large-scale models into semantic blocks and enabling FL participants to opportunistically upload selected important blocks in a quantized manner, it significantly reduces the communication overhead while maintaining model performance. Since its deployment in ENN Group in February 2022, FEDOBD has served two coal chemical plants across two cities in China to build industrial fault prediction models. It helped the company reduce the training communication overhead by over 70% compared to its previous AI Engine, while maintaining model performance at over 85% test F1 score. To our knowledge, it is the first successfully deployed dropout-based FL approach.

Random Fourier Features (RFF) is among the most popular and broadly applicable approaches for scaling up kernel methods. In essence, RFF allows the user to avoid costly computations on a large kernel matrix via a fast randomized approximation. However, a pervasive difficulty in applying RFF is that the user does not know the actual error of the approximation, or how this error will propagate into downstream learning tasks. Up to now, the RFF literature has primarily dealt with these uncertainties using theoretical error bounds, but from a user's standpoint, such results are typically impractical -- either because they are highly conservative or involve unknown quantities. To tackle these general issues in a data-driven way, this paper develops a bootstrap approach to numerically estimate the errors of RFF approximations. Three key advantages of this approach are: (1) The error estimates are specific to the problem at hand, avoiding the pessimism of worst-case bounds. (2) The approach is flexible with respect to different uses of RFF, and can even estimate errors in downstream learning tasks. (3) The approach enables adaptive computation, so that the user can quickly inspect the error of a rough initial kernel approximation and then predict how much extra work is needed. Lastly, in exchange for all of these benefits, the error estimates can be obtained at a modest computational cost.

In this technical report, we present our solution for the Baidu KDD Cup 2022 Spatial Dynamic Wind Power Forecasting Challenge. Wind power is a rapidly growing source of clean energy. Accurate wind power forecasting is essential for grid stability and the security of supply. Therefore, organizers provide a wind power dataset containing historical data from 134 wind turbines and launch the Baidu KDD Cup 2022 to examine the limitations of current methods for wind power forecasting. The average of RMSE (Root Mean Square Error) and MAE (Mean Absolute Error) is used as the evaluation score. We adopt two spatial-temporal graph neural network models, i.e., AGCRN and MTGNN, as our basic models. We train AGCRN by 5-fold cross-validation and additionally train MTGNN directly on the training and validation sets. Finally, we ensemble the two models based on the loss values of the validation set as our final submission. Using our method, our team \team achieves -45.36026 on the test set. We release our codes on Github (https://github.com/BUAABIGSCity/KDDCUP2022) for reproduction.