We present Sandwich Batch Normalization (SaBN), an embarrassingly easy improvement of Batch Normalization (BN) with only a few lines of code changes. SaBN is motivated by addressing the inherent feature distribution heterogeneity that one can be identified in many tasks, which can arise from data heterogeneity (multiple input domains) or model heterogeneity (dynamic architectures, model conditioning, etc.). Our SaBN factorizes the BN affine layer into one shared sandwich affine layer, cascaded by several parallel independent affine layers. Concrete analysis reveals that, during optimization, SaBN promotes balanced gradient norms while still preserving diverse gradient directions: a property that many application tasks seem to favor. We demonstrate the prevailing effectiveness of SaBN as a drop-in replacement in four tasks: $\textbf{conditional image generation}$, $\textbf{neural architecture search}$ (NAS), $\textbf{adversarial training}$, and $\textbf{arbitrary style transfer}$. Leveraging SaBN immediately achieves better Inception Score and FID on CIFAR-10 and ImageNet conditional image generation with three state-of-the-art GANs; boosts the performance of a state-of-the-art weight-sharing NAS algorithm significantly on NAS-Bench-201; substantially improves the robust and standard accuracies for adversarial defense; and produces superior arbitrary stylized results. We also provide visualizations and analysis to help understand why SaBN works. Codes are available at https://github.com/VITA-Group/Sandwich-Batch-Normalization.

Lines are human mental abstractions. A bunch of lines may form a drawing. A set of drawings can feed an LSTM network input layer, considering each draw as a list of lines and a line a list of points. This paper proposes the pointless motive to classify the gender of celebrities' portraits as an excuse for exploration in a broad, more artistic sense. Investigation results drove compelling ideas here discussed. The experiments compared different ways to represent draws to be input in a network and showed that an absolute format of coordinates (x, y) was a better performer than a relative one (Dx, Dy) with respect to prior points, most frequent in the reviewed literature. Experiments also showed that, due to the recurrent nature of LSTMs, the order of lines forming a drawing is a relevant factor for input in an LSTM classifier not studied before. A minimum 'pencil' traveled length criteria for line ordering proved suitable, possible by reducing it to a TSP particular instance. The best configuration for gender classification appears with an LSTM layer that returns the hidden state value for each input point step, followed by a global average layer along the sequence, before the output dense layer. That result guided the idea of removing the average in the network pipeline and return a per-point attribute score just by adjusting tensors dimensions. With this trick, the model detects an attribute in a drawing and also recognizes the points linked to it. Moreover, by overlapping filtered lines of portraits, an attribute's visual essence is depicted. Meet the FaCells.

Learning effective representations in image-based environments is crucial for sample efficient Reinforcement Learning (RL). Unfortunately, in RL, representation learning is confounded with the exploratory experience of the agent -- learning a useful representation requires diverse data, while effective exploration is only possible with coherent representations. Furthermore, we would like to learn representations that not only generalize across tasks but also accelerate downstream exploration for efficient task-specific training. To address these challenges we propose Proto-RL, a self-supervised framework that ties representation learning with exploration through prototypical representations. These prototypes simultaneously serve as a summarization of the exploratory experience of an agent as well as a basis for representing observations. We pre-train these task-agnostic representations and prototypes on environments without downstream task information. This enables state-of-the-art downstream policy learning on a set of difficult continuous control tasks.

Automatic essay grading (AEG) is a process in which machines assign a grade to an essay written in response to a topic, called the prompt. Zero-shot AEG is when we train a system to grade essays written to a new prompt which was not present in our training data. In this paper, we describe a solution to the problem of zero-shot automatic essay grading, using cognitive information, in the form of gaze behaviour. Our experiments show that using gaze behaviour helps in improving the performance of AEG systems, especially when we provide a new essay written in response to a new prompt for scoring, by an average of almost 5 percentage points of QWK.

Image Captioning, or the automatic generation of descriptions for images, is one of the core problems in Computer Vision and has seen considerable progress using Deep Learning Techniques. We propose to use Inception-ResNet Convolutional Neural Network as encoder to extract features from images, Hierarchical Context based Word Embeddings for word representations and a Deep Stacked Long Short Term Memory network as decoder, in addition to using Image Data Augmentation to avoid over-fitting. For data Augmentation, we use Horizontal and Vertical Flipping in addition to Perspective Transformations on the images. We evaluate our proposed methods with two image captioning frameworks- Encoder-Decoder and Soft Attention. Evaluation on widely used metrics have shown that our approach leads to considerable improvement in model performance.

The recent success of supervised deep learning is built upon two crucial cornerstones: That the training and test data are drawn from an identical distribution, and that representative labeled data are available for training. However, in real-world applications, labeled data drawn from the same distribution as test data are usually unavailable. Domain adaptation (Quionero-Candela et al., 2009; Saenko et al., 2010) suggests a way to overcome this challenge by transferring the knowledge of labeled data from a source domain to the target domain. Without further assumptions, the transferability of information is not possible. Existing theoretical works have investigated suitable assumptions that can provide learning guarantees. Many of the works are based on the covariate shift assumption (Heckman, 1979; Shimodaira, 2000), which states that the conditional distribution of the labels (given the input x) is invariant across domains, i.e., p

Deep learning models suffer from opaqueness. For Convolutional Neural Networks (CNNs), current research strategies for explaining models focus on the target classes within the associated training dataset. As a result, the understanding of hidden feature map activations is limited by the discriminative knowledge gleaned during training. The aim of our work is to explain and expand CNNs models via the mirroring or alignment of CNN to an external knowledge base. This will allow us to give a semantic context or label for each visual feature. We can match CNN feature activations to nodes in our external knowledge base. This supports knowledge-based interpretation of the features associated with model decisions. To demonstrate our approach, we build two separate graphs. We use an entity alignment method to align the feature nodes in a CNN with the nodes in a ConceptNet based knowledge graph. We then measure the proximity of CNN graph nodes to semantically meaningful knowledge base nodes. Our results show that in the aligned embedding space, nodes from the knowledge graph are close to the CNN feature nodes that have similar meanings, indicating that nodes from an external knowledge base can act as explanatory semantic references for features in the model. We analyse a variety of graph building methods in order to improve the results from our embedding space. We further demonstrate that by using hierarchical relationships from our external knowledge base, we can locate new unseen classes outside the CNN training set in our embeddings space, based on visual feature activations. This suggests that we can adapt our approach to identify unseen classes based on CNN feature activations. Our demonstrated approach of aligning a CNN with an external knowledge base paves the way to reason about and beyond the trained model, with future adaptations to explainable models and zero-shot learning.

The two fields of machine learning and graphical causality arose and developed separately. However, there is now cross-pollination and increasing interest in both fields to benefit from the advances of the other. In the present paper, we review fundamental concepts of causal inference and relate them to crucial open problems of machine learning, including transfer and generalization, thereby assaying how causality can contribute to modern machine learning research. This also applies in the opposite direction: we note that most work in causality starts from the premise that the causal variables are given. A central problem for AI and causality is, thus, causal representation learning, the discovery of high-level causal variables from low-level observations. Finally, we delineate some implications of causality for machine learning and propose key research areas at the intersection of both communities.

Transformers are arguably the main workhorse in recent Natural Language Processing research. By definition a Transformer is invariant with respect to reorderings of the input. However, language is inherently sequential and word order is essential to the semantics and syntax of an utterance. In this paper, we provide an overview of common methods to incorporate position information into Transformer models. The objectives of this survey are to i) showcase that position information in Transformer is a vibrant and extensive research area; ii) enable the reader to compare existing methods by providing a unified notation and meaningful clustering; iii) indicate what characteristics of an application should be taken into account when selecting a position encoding; iv) provide stimuli for future research. The Transformer model as introduced by Vaswani et al. (2017) has been found to perform well for many tasks, such as machine translation or language modeling. With the rise of pretrained language models (PLMs) (Peters et al., 2018; Howard & Ruder, 2018; Devlin et al., 2019; Brown et al., 2020) Transformer models have become even more popular. As a result they are at the core of many state of the art natural language processing (NLP) models. A Transformer model consists of several layers, or blocks. Each layer is a self-attention (Vaswani et al., 2017) module followed by a feed-forward layer. Layer normalization and residual connections are additional components of a layer.

We study the problem of incorporating prior knowledge into a deep Transformer-based model,i.e.,Bidirectional Encoder Representations from Transformers (BERT), to enhance its performance on semantic textual matching tasks. By probing and analyzing what BERT has already known when solving this task, we obtain better understanding of what task-specific knowledge BERT needs the most and where it is most needed. The analysis further motivates us to take a different approach than most existing works. Instead of using prior knowledge to create a new training task for fine-tuning BERT, we directly inject knowledge into BERT's multi-head attention mechanism. This leads us to a simple yet effective approach that enjoys fast training stage as it saves the model from training on additional data or tasks other than the main task. Extensive experiments demonstrate that the proposed knowledge-enhanced BERT is able to consistently improve semantic textual matching performance over the original BERT model, and the performance benefit is most salient when training data is scarce.