Transfer learning where the behavior of extracting transferable knowledge from the source domain(s) and reusing this knowledge to target domain has become a research area of great interest in the field of artificial intelligence. Probabilistic graphical models (PGMs) have been recognized as a powerful tool for modeling complex systems with many advantages, e.g., the ability to handle uncertainty and possessing good interpretability. Considering the success of these two aforementioned research areas, it seems natural to apply PGMs to transfer learning. However, although there are already some excellent PGMs specific to transfer learning in the literature, the potential of PGMs for this problem is still grossly underestimated. This paper aims to boost the development of PGMs for transfer learning by 1) examining the pilot studies on PGMs specific to transfer learning, i.e., analyzing and summarizing the existing mechanisms particularly designed for knowledge transfer; 2) discussing examples of real-world transfer problems where existing PGMs have been successfully applied; and 3) exploring several potential research directions on transfer learning using PGM.

We develop a Distributionally Robust Optimization (DRO) formulation for Multiclass Logistic Regression (MLR), which could tolerate data contaminated by outliers. The DRO framework uses a probabilistic ambiguity set defined as a ball of distributions that are close to the empirical distribution of the training set in the sense of the Wasserstein metric. We relax the DRO formulation into a regularized learning problem whose regularizer is a norm of the coefficient matrix. We establish out-of-sample performance guarantees for the solutions to our model, offering insights on the role of the regularizer in controlling the prediction error. We apply the proposed method in rendering deep CNN-based image classifiers robust to random and adversarial attacks. Specifically, using the MNIST and CIFAR-10 datasets, we demonstrate reductions in test error rate by up to 78.8% and loss by up to 90.8%. We also show that with a limited number of perturbed images in the training set, our method can improve the error rate by up to 49.49% and the loss by up to 68.93% compared to Empirical Risk Minimization (ERM), converging faster to an ideal loss/error rate as the number of perturbed images increases.

We propose Hypernetwork Kalman Filter (HKF) for tracking applications with multiple different dynamics. The HKF combines generalization power of Kalman filters with expressive power of neural networks. Instead of keeping a bank of Kalman filters and choosing one based on approximating the actual dynamics, HKF adapts itself to each dynamics based on the observed sequence. Through extensive experiments on CDL-B channel model, we show that the HKF can be used for tracking the channel over a wide range of Doppler values, matching Kalman filter performance with genie Doppler information. At high Doppler values, it achieves around 2dB gain over genie Kalman filter. The HKF generalizes well to unseen Doppler, SNR values and pilot patterns unlike LSTM, which suffers from severe performance degradation.

The increasing availability of massive data sets poses a series of challenges for machine learning. Prominent among these is the need to learn models under hardware or human resource constraints. In such resource-constrained settings, a simple yet powerful approach is to operate on small subsets of the data. Coresets are weighted subsets of the data that provide approximation guarantees for the optimization objective. However, existing coreset constructions are highly model-specific and are limited to simple models such as linear regression, logistic regression, and $k$-means. In this work, we propose a generic coreset construction framework that formulates the coreset selection as a cardinality-constrained bilevel optimization problem. In contrast to existing approaches, our framework does not require model-specific adaptations and applies to any twice differentiable model, including neural networks. We show the effectiveness of our framework for a wide range of models in various settings, including training non-convex models online and batch active learning.

Forecasting the trajectories of neighbor vehicles is a crucial step for decision making and motion planning of autonomous vehicles. This paper proposes a graph-based spatial-temporal convolutional network (GSTCN) to predict future trajectory distributions of all neighbor vehicles using past trajectories. This network tackles the spatial interactions using a graph convolutional network (GCN), and captures the temporal features with a convolutional neural network (CNN). The spatial-temporal features are encoded and decoded by a gated recurrent unit (GRU) network to generate future trajectory distributions. Besides, we propose a weighted adjacency matrix to describe the intensities of mutual influence between vehicles, and the ablation study demonstrates the effectiveness of our proposed scheme. Our network is evaluated on two real-world freeway trajectory datasets: I-80 and US-101 in the Next Generation Simulation (NGSIM).Comparisons in three aspects, including prediction errors, model sizes, and inference speeds, show that our network can achieve state-of-the-art performance.

The ability to abstract, count, and use System 2 reasoning are well-known manifestations of intelligence and understanding. In this paper, we argue, using the example of the ``Look and Say" puzzle, that although deep neural networks can exhibit high `competence' (as measured by accuracy) when trained on large data sets (2M examples in our case), they do not show any sign on the deeper understanding of the problem, or what D. Dennett calls `comprehension'. We report on two sets experiments on the ``Look and Say" puzzle data. We view the problem as building a translator from one set of tokens to another. We apply both standard LSTMs and Transformer/Attention -- based neural networks, using publicly available machine translation software. We observe that despite the amazing accuracy (on both, training and test data), the performance of the trained programs on the actual L\&S sequence is bad. We then discuss a few possible ramifications of this finding and connections to other work, experimental and theoretical. First, from the cognitive science perspective, we argue that we need better mathematical models of abstraction. Second, the classical and more recent results on the universality of neural networks should be re-examined for functions acting on discrete data sets. Mapping on discrete sets usually have no natural continuous extensions. This connects the results on a simple puzzle to more sophisticated results on modeling of mathematical functions, where algebraic functions are more difficult to model than e.g. differential equations. Third, we hypothesize that for problems such as ``Look and Say", computing the parity of bitstrings, or learning integer addition, it might be worthwhile to introduce concepts from topology, where continuity is defined without the reference to the concept of distance.

Facial expression recognition is a challenging task when neural network is applied to pattern recognition. Most of the current recognition research is based on single source facial data, which generally has the disadvantages of low accuracy and low robustness. In this paper, a neural network algorithm of facial expression recognition based on multimodal data fusion is proposed. The algorithm is based on the multimodal data, and it takes the facial image, the histogram of oriented gradient of the image and the facial landmarks as the input, and establishes CNN, LNN and HNN three sub neural networks to extract data features, using multimodal data feature fusion mechanism to improve the accuracy of facial expression recognition. Experimental results show that, benefiting by the complementarity of multimodal data, the algorithm has a great improvement in accuracy, robustness and detection speed compared with the traditional facial expression recognition algorithm. Especially in the case of partial occlusion, illumination and head posture transformation, the algorithm also shows a high confidence.

Recently, multiple approaches for creating agents for playing various complex real-time computer games such as StarCraft II or Dota 2 were proposed, however, they either embed a significant amount of expert knowledge into the agent or use a prohibitively large for most researchers amount of computational resources. We propose a neural network architecture for playing the full two-player match of StarCraft II trained with general-purpose supervised and reinforcement learning, that can be trained on a single consumer-grade PC with a single GPU. We also show that our implementation achieves a non-trivial performance when compared to the in-game scripted bots. We make no simplifying assumptions about the game except for playing on a single chosen map, and we use very little expert knowledge. In principle, our approach can be applied to any RTS game with small modifications. While our results are far behind the state-of-the-art large-scale approaches in terms of the final performance, we believe our work can serve as a solid baseline for other small-scale experiments.

There is a soaring interest in the news recommendation research scenario due to the information overload. To accurately capture users' interests, we propose to model multi-modal features, in addition to the news titles that are widely used in existing works, for news recommendation. Besides, existing research pays little attention to the click decision-making process in designing multi-modal modeling modules. In this work, inspired by the fact that users make their click decisions mostly based on the visual impression they perceive when browsing news, we propose to capture such visual impression information with visual-semantic modeling for news recommendation. Specifically, we devise the local impression modeling module to simultaneously attend to decomposed details in the impression when understanding the semantic meaning of news title, which could explicitly get close to the process of users reading news. In addition, we inspect the impression from a global view and take structural information, such as the arrangement of different fields and spatial position of different words on the impression, into the modeling of multiple modalities. To accommodate the research of visual impression-aware news recommendation, we extend the text-dominated news recommendation dataset MIND by adding snapshot impression images and will release it to nourish the research field. Extensive comparisons with the state-of-the-art news recommenders along with the in-depth analyses demonstrate the effectiveness of the proposed method and the promising capability of modeling visual impressions for the content-based recommenders.

In this paper we introduce the Local Logo Generative Adversarial Network (LL-GAN) that uses regional features extracted from the Faster Regional Convolutional Neural Network (Faster R-CNN) to generate logos. We demonstrate the strength of this approach by training the framework on a small style-rich dataset collected online to generate large impressive logos. Our approach beats the state-of-the-art models (StyleGAN2, Self-Attention GANs) that suffer from mode collapse due to the size of the data.