Multi-objective optimization is key to solving many Engineering Design problems, where design parameters are optimized for several performance indicators. However, optimization results are highly dependent on how the designs are parameterized. Researchers have shown that deep generative models can learn compact design representations, providing a new way of parameterizing designs to achieve faster convergence and improved optimization performance. Despite their success in capturing complex distributions, existing generative models face three challenges when used for design problems: 1) generated designs have limited design space coverage, 2) the generator ignores design performance, and 3) the new parameterization is unable to represent designs beyond training data. To address these challenges, we propose MO-PaDGAN, which adds a Determinantal Point Processes based loss function to the generative adversarial network to simultaneously model diversity and (multi-variate) performance. MO-PaDGAN can thus improve the performances and coverage of generated designs, and even generate designs with performances exceeding those from training data. When using MO-PaDGAN as a new parameterization in multi-objective optimization, we can discover much better Pareto fronts even though the training data do not cover those Pareto fronts. In a real-world multi-objective airfoil design example, we demonstrate that MO-PaDGAN achieves, on average, a 186% improvement in the hypervolume indicator when compared to the vanilla GAN or other state-of-the-art parameterization methods.

This article presents our unimodal privacy-safe and non-individual proposal for the audio-video group emotion recognition subtask at the Emotion Recognition in the Wild (EmotiW) Challenge 2020 1. This sub challenge aims to classify in the wild videos into three categories: Positive, Neutral and Negative. Recent deep learning models have shown tremendous advances in analyzing interactions between people, predicting human behavior and affective evaluation. Nonetheless, their performance comes from individual-based analysis, which means summing up and averaging scores from individual detections, which inevitably leads to some privacy issues. In this research, we investigated a frugal approach towards a model able to capture the global moods from the whole image without using face or pose detection, or any individual-based feature as input. The proposed methodology mixes state-of-the-art and dedicated synthetic corpora as training sources. With an in-depth exploration of neural network architectures for group-level emotion recognition, we built a VGG-based model achieving 59.13% accuracy on the VGAF test set (eleventh place of the challenge). Given that the analysis is unimodal based only on global features and that the performance is evaluated on a real-world dataset, these results are promising and let us envision extending this model to multimodality for classroom ambiance evaluation, our final target application.

Unsupervised (or self-supervised) graph representation learning is essential to facilitate various graph data mining tasks when external supervision is unavailable. The challenge is to encode the information about the graph structure and the attributes associated with the nodes and edges into a low dimensional space. Most existing unsupervised methods promote similar representations across nodes that are topologically close. Recently, it was shown that leveraging additional graph-level information, e.g., information that is shared among all nodes, encourages the representations to be mindful of the global properties of the graph, which greatly improves their quality. However, in most graphs, there is significantly more structure that can be captured, e.g., nodes tend to belong to (multiple) clusters that represent structurally similar nodes. Motivated by this observation, we propose a graph representation learning method called Graph InfoClust (GIC), that seeks to additionally capture cluster-level information content. These clusters are computed by a differentiable K-means method and are jointly optimized by maximizing the mutual information between nodes of the same clusters. This optimization leads the node representations to capture richer information and nodal interactions, which improves their quality. Experiments show that GIC outperforms state-of-art methods in various downstream tasks (node classification, link prediction, and node clustering) with a 0.9% to 6.1% gain over the best competing approach, on average.

Physics-Informed Neural Networks (PINNs) have emerged recently as a promising application of deep neural networks to the numerical solution of nonlinear partial differential equations (PDEs). However, the original PINN algorithm is known to suffer from stability and accuracy problems in cases where the solution has sharp spatio-temporal transitions. These stiff PDEs require an unreasonably large number of collocation points to be solved accurately. It has been recognized that adaptive procedures are needed to force the neural network to fit accurately the stubborn spots in the solution of stiff PDEs. To accomplish this, previous approaches have used fixed weights hard-coded over regions of the solution deemed to be important. In this paper, we propose a fundamentally new method to train PINNs adaptively, where the adaptation weights are fully trainable, so the neural network learns by itself which regions of the solution are difficult and is forced to focus on them, which is reminiscent of soft multiplicative-mask attention mechanism used in computer vision. The basic idea behind these Self-Adaptive PINNs is to make the weights increase where the corresponding loss is higher, which is accomplished by training the network to simultaneously minimize the losses and maximize the weights, i.e., to find a saddle point in the cost surface. We show that this is formally equivalent to solving a PDE-constrained optimization problem using a penalty-based method, though in a way where the monotonically-nondecreasing penalty coefficients are trainable. Numerical experiments with an Allen-Cahn stiff PDE, the Self-Adaptive PINN outperformed other state-of-the-art PINN algorithms in L2 error by a wide margin, while using a smaller number of training epochs. An Appendix contains additional results with Burger's and Helmholtz PDEs, which confirmed the trends observed in the Allen-Cahn experiments.

The core operation of current Graph Neural Networks (GNNs) is the aggregation enabled by the graph Laplacian or message passing, which filters the neighborhood node information. Though effective for various tasks, in this paper, we show that they are potentially a problematic factor underlying all GNN methods for learning on certain datasets, as they force the node representations similar, making the nodes gradually lose their identity and become indistinguishable. Hence, we augment the aggregation operations with their dual, i.e. diversification operators that make the node more distinct and preserve the identity. Such augmentation replaces the aggregation with a two-channel filtering process that, in theory, is beneficial for enriching the node representations. In practice, the proposed two-channel filters can be easily patched on existing GNN methods with diverse training strategies, including spectral and spatial (message passing) methods. In the experiments, we observe desired characteristics of the models and significant performance boost upon the baselines on 9 node classification tasks.

Originated from distributed learning, federated learning enables privacy-preserved collaboration on a new abstracted level by sharing the model parameters only. While the current research mainly focuses on optimizing learning algorithms and minimizing communication overhead left by distributed learning, there is still a considerable gap when it comes to the real implementation on mobile devices. In this paper, we start with an empirical experiment to demonstrate computation heterogeneity is a more pronounced bottleneck than communication on the current generation of battery-powered mobile devices, and the existing methods are haunted by mobile stragglers. Further, non-identically distributed data across the mobile users makes the selection of participants critical to the accuracy and convergence. To tackle the computational and statistical heterogeneity, we utilize data as a tuning knob and propose two efficient polynomial-time algorithms to schedule different workloads on various mobile devices, when data is identically or non-identically distributed. For identically distributed data, we combine partitioning and linear bottleneck assignment to achieve near-optimal training time without accuracy loss. For non-identically distributed data, we convert it into an average cost minimization problem and propose a greedy algorithm to find a reasonable balance between computation time and accuracy. We also establish an offline profiler to quantify the runtime behavior of different devices, which serves as the input to the scheduling algorithms. We conduct extensive experiments on a mobile testbed with two datasets and up to 20 devices. Compared with the common benchmarks, the proposed algorithms achieve 2-100x speedup epoch-wise, 2-7% accuracy gain and boost the convergence rate by more than 100% on CIFAR10.

In this work, we focus on large graph similarity computation problem and propose a novel "embedding-coarsening-matching" learning framework, which outperforms state-of-the-art methods in this task and has significant improvement in time efficiency. Graph similarity computation for metrics such as Graph Edit Distance (GED) is typically NP-hard, and existing heuristics-based algorithms usually achieves a unsatisfactory trade-off between accuracy and efficiency. Recently the development of deep learning techniques provides a promising solution for this problem by a data-driven approach which trains a network to encode graphs to their own feature vectors and computes similarity based on feature vectors. These deep-learning methods can be classified to two categories, embedding models and matching models. Embedding models such as GCN-Mean and GCN-Max, which directly map graphs to respective feature vectors, run faster but the performance is usually poor due to the lack of interactions across graphs. Matching models such as GMN, whose encoding process involves interaction across the two graphs, are more accurate but interaction between whole graphs brings a significant increase in time consumption (at least quadratic time complexity over number of nodes). Inspired by large biological molecular identification where the whole molecular is first mapped to functional groups and then identified based on these functional groups, our "embedding-coarsening-matching" learning framework first embeds and coarsens large graphs to coarsened graphs with denser local topology and then matching mechanism is deployed on the coarsened graphs for the final similarity scores. Detailed experiments have been conducted and the results demonstrate the efficiency and effectiveness of our proposed framework.

Due to their susceptibility to adversarial perturbations, neural networks (NNs) are hardly used in safety-critical applications. One measure of robustness to such perturbations in the input is the Lipschitz constant of the input-output map defined by an NN. In this work, we propose a framework to train multi-layer NNs while at the same time encouraging robustness by keeping their Lipschitz constant small, thus addressing the robustness issue. More specifically, we design an optimization scheme based on the Alternating Direction Method of Multipliers that minimizes not only the training loss of an NN but also its Lipschitz constant resulting in a semidefinite programming based training procedure that promotes robustness. We design two versions of this training procedure. The first one includes a regularizer that penalizes an accurate upper bound on the Lipschitz constant. The second one allows to enforce a desired Lipschitz bound on the NN at all times during training. Finally, we provide two examples to show that the proposed framework successfully increases the robustness of NNs.

In modern businesses, it really depends more on the overhead if you ask me. Can you buy a gpu easily through your it department? Can you schedule a cloud gpu easily internally? Can you start/stop it yourself or are those semi-admin rights not given to you and do you need to get approval every time (this happens more than you'd think...) Can you run a Linux environment or does your it require you to use windows if you go the desktop option?

How much knowledge do pretrained language models hold? Recent research observed that pretrained transformers are adept at modeling semantics but it is unclear to what degree they grasp human knowledge, or how to ensure they do so. In this paper we incorporate knowledge-awareness in language model pretraining without changing the transformer architecture, inserting explicit knowledge layers, or adding external storage of semantic information. Rather, we simply signal the existence of entities to the input of the transformer in pretraining, with an entity-extended tokenizer; and at the output, with an additional entity prediction task. Our experiments show that solely by adding these entity signals in pretraining, significantly more knowledge is packed into the transformer parameters: we observe improved language modeling accuracy, factual correctness in LAMA knowledge probing tasks, and semantics in the hidden representations through edge probing.