Analysis of large-scale sequential data has been one of the most crucial tasks in areas such as bioinformatics, text, and audio mining. Existing string kernels, however, either (i) rely on local features of short substructures in the string, which hardly capture long discriminative patterns, (ii) sum over too many substructures, such as all possible subsequences, which leads to diagonal dominance of the kernel matrix, or (iii) rely on non-positive-definite similarity measures derived from the edit distance. Furthermore, while there have been works addressing the computational challenge with respect to the length of string, most of them still experience quadratic complexity in terms of the number of training samples when used in a kernel-based classifier. In this paper, we present a new class of global string kernels that aims to (i) discover global properties hidden in the strings through global alignments, (ii) maintain positive-definiteness of the kernel, without introducing a diagonal dominant kernel matrix, and (iii) have a training cost linear with respect to not only the length of the string but also the number of training string samples. To this end, the proposed kernels are explicitly defined through a series of different random feature maps, each corresponding to a distribution of random strings. We show that kernels defined this way are always positive-definite, and exhibit computational benefits as they always produce \emph{Random String Embeddings (RSE)} that can be directly used in any linear classification models. Our extensive experiments on nine benchmark datasets corroborate that RSE achieves better or comparable accuracy in comparison to state-of-the-art baselines, especially with the strings of longer lengths. In addition, we empirically show that RSE scales linearly with the increase of the number and the length of string.

Charge prediction which aims to determine appropriate charges for criminal cases based on textual fact descriptions, is an important technology in the field of AI&Law. Previous works focus on improving prediction accuracy, ignoring the interpretability, which limits the methods' applicability. In this work, we propose a deep neural framework to extract short but charge-decisive text snippets - rationales - from input fact description, as the interpretation of charge prediction. To solve the scarcity problem of rationale annotated corpus, rationales are extracted in a reinforcement style with the only supervision in the form of charge labels. We further propose a dynamic rationale attention mechanism to better utilize the information in extracted rationales and predict the charges. Experimental results show that besides providing charge prediction interpretation, our approach can also capture subtle details to help charge prediction.

Pre-Training of Deep Bidirectional Protein Sequence Representations with Structural Information Seonwoo Min, 1 Seunghyun Park, 2 Siwon Kim, 1 Hyun-Soo Choi, 1 Sungroh Y oon 1, 3, † 1 Department of Electrical and Computer Engineering, Seoul National University, Seoul 08826, Korea 2 Clova AI Research, NA VER Corp., Seongnam 13561, Korea 3 Interdisciplinary Program in Bioinformatics, ASRI, INMC, and ISRC, Seoul National University, Seoul 08826, Korea † Correspondence to: sryoon@snu.ac.kr Abstract A structure of a protein has a direct impact on its properties and functions. However, identification of structural similarity directly from amino acid sequences remains as a challenging problem in computational biology. In this paper, we introduce a novel BERT -wise pre-training scheme for a protein sequence representation model called PLUS, which stands for Protein sequence representations L earned U sing Structural information. As natural language representation models capture syntactic and semantic information of words from a large unlabeled text corpus, PLUS captures structural information of amino acids from a large weakly labeled protein database. Since the Transformer encoder, BERT's original model architecture, has a severe computational requirement to handle long sequences, we first propose to combine a bidirectional recurrent neural network with the BERT -wise pre-training scheme. PLUS is designed to learn protein representations with two pre-training objectives, i.e., masked language modeling and same family prediction. Then, the pre-trained model can be fine-tuned for a wide range of tasks without training randomly initialized task-specific models from scratch. Introduction Proteins consisting of linear chains of amino acids are the most versatile molecules in living organisms. They serve vital functions in almost every biological mechanism, e.g., transmitting nerve pulses, storing and transporting other molecules, and providing immune protection (Berg, Ty-moczko, and Stryer 2006).

This paper tackles the goal of conclusion-supplement answer generation for non-factoid questions, which is a critical issue in the field of Natural Language Processing (NLP) and Artificial Intelligence (AI), as users often require supplementary information before accepting a conclusion. The current encoder-decoder framework, however, has difficulty generating such answers, since it may become confused when it tries to learn several different long answers to the same non-factoid question. Our solution, called an ensemble network, goes beyond single short sentences and fuses logically connected conclusion statements and supplementary statements. It extracts the context from the conclusion decoder's output sequence and uses it to create supplementary decoder states on the basis of an attention mechanism. It also assesses the closeness of the question encoder's output sequence and the separate outputs of the conclusion and supplement decoders as well as their combination. As a result, it generates answers that match the questions and have natural-sounding supplementary sequences in line with the context expressed by the conclusion sequence. Evaluations conducted on datasets including "Love Advice" and "Arts & Humanities" categories indicate that our model outputs much more accurate results than the tested baseline models do.

We present a deep neural network-based methodology for synthesising percussive sounds with control over high-level timbral characteristics of the sounds. This approach allows for intuitive control of a synthesizer, enabling the user to shape sounds without extensive knowledge of signal processing. We use a feedforward convolutional neural network-based architecture, which is able to map input parameters to the corresponding waveform. We propose two datasets to evaluate our approach on both a restrictive context, and in one covering a broader spectrum of sounds. The timbral features used as parameters are taken from recent literature in signal processing. We also use these features for evaluation and validation of the presented model, to ensure that changing the input parameters produces a congruent waveform with the desired characteristics. Finally, we evaluate the quality of the output sound using a subjective listening test. We provide sound examples and the system's source code for reproducibility.

Deep reinforcement learning requires a heavy price in terms of sample efficiency and overparameterization in the neural networks used for function approximation. In this work, we use tensor factorization in order to learn more compact representation for reinforcement learning policies. We show empirically that in the low-data regime, it is possible to learn online policies with 2 to 10 times less total coefficients, with little to no loss of performance. We also leverage progress in second order optimization, and use the theory of wavelet scattering to further reduce the number of learned coefficients, by foregoing learning the topmost convolutional layer filters altogether. We evaluate our results on the Atari suite against recent baseline algorithms that represent the state-of-the-art in data efficiency, and get comparable results with an order of magnitude gain in weight parsimony.

--Inspired by the human visual perception system, hexagonal image processing in the context of machine learning deals with the development of image processing systems that combine the advantages of evolutionary motivated structures based on biological models. While conventional state of the art image processing systems of recording and output devices almost exclusively utilize square arranged methods, their hexagonal counterparts offer a number of key advantages that can benefit both researchers and users. This contribution serves as a general application-oriented approach with the synthesis of the therefor designed hexagonal image processing framework, called Hexnet, the processing steps of hexagonal image transformation, and dependent methods. The results of our created test environment show that the realized framework surpasses current approaches of hexagonal image processing systems, while hexagonal artificial neural networks can benefit from the implemented hexagonal architecture. As hexagonal lattice format based deep neural networks, also called H-DNN, can be compared to their square counterpart by transforming classical square lattice based data sets into their hexagonal representation, they can also result in a reduction of trainable parameters as well as result in increased training and test rates.

Automated visual inspection in the semiconductor industry aims to detect and classify manufacturing defects utilizing modern image processing techniques. While an earliest possible detection of defect patterns allows quality control and automation of manufacturing chains, manufacturers benefit from an increased yield and reduced manufacturing costs. Since classical image processing systems are limited in their ability to detect novel defect patterns, and machine learning approaches often involve a tremendous amount of computational effort, this contribution introduces a novel deep neural network-based hybrid approach. Unlike classical deep neural networks, a multi-stage system allows the detection and classification of the finest structures in pixel size within high-resolution imagery. Consisting of stacked hybrid convolutional neural networks (SH-CNN) and inspired by current approaches of visual attention, the realized system draws the focus over the level of detail from its structures to more task-relevant areas of interest. The results of our test environment show that the SH-CNN outperforms current approaches of learning-based automated visual inspection, whereas a distinction depending on the level of detail enables the elimination of defect patterns in earlier stages of the manufacturing process.

The uncertainty estimation is critical in real-world decision making applications, especially when distributional shift between the training and test data are prevalent. Many calibration methods in the literature have been proposed to improve the predictive uncertainty of DNNs which are generally not well-calibrated. However, none of them is specifically designed to work properly under domain shift condition. In this paper, we propose Unsupervised Temperature Scaling (UTS) as a robust calibration method to domain shift. It exploits unlabeled test samples instead of the training one to adjust the uncertainty prediction of deep models towards the test distribution. UTS utilizes a novel loss function, weighted NLL, which allows unsupervised calibration. We evaluate UTS on a wide range of model-datasets to show the possibility of calibration without labels and demonstrate the robustness of UTS compared to other methods (e.g., TS, MCdropout, SVI, ensembles) in shifted domains. The predictive distributions provided by Deep Neural Networks (DNNs) have been increasingly used for decision-support systems, for applications ranging from medical diagnoses assistance (Esteva et al., 2017) to self-driving cars (Bojarski et al., 2016). In DNNs, the predictive distributions usually corresponds to the output of a softmax layer, which is typically interpreted as the confidence over the different classes. The i.i.d hypothesis made in learning usually assumes that the data distributions over the classes are the same at learning and inference time. However, in real-world applications, the distribution of data at inference time (i.e., the test data) may shift and actually be different from the original training distribution - corresponding to distribution shift in representation of data which we refer that as domain shift. For instance, in image classification problem, domain shift happens when the test images are different in illumination, view point, resolution, background or intensity noise from the training set. However, they are the same classification problem with the same objects occurance rate.

Despite the success of deep neural networks (DNNs), state-of-the-art models are too large to deploy on low-resource devices or common server configurations in which multiple models are held in memory. Model compression methods address this limitation by reducing the memory footprint, latency, or energy consumption of a model with minimal impact on accuracy. We focus on the task of reducing the number of learnable variables in the model. In this work we combine ideas from weight hashing and dimensionality reductions resulting in a simple and powerful structured multi-hashing method based on matrix products that allows direct control of model size of any deep network and is trained end-to-end. We demonstrate the strength of our approach by compressing models from the ResNet, EfficientNet, and MobileNet architecture families. Our method allows us to drastically decrease the number of variables while maintaining high accuracy. For instance, by applying our approach to EfficentNet-B4 (16M parameters) we reduce it to to the size of B0 (5M parameters), while gaining over 3% in accuracy over B0 baseline. On the commonly used benchmark CIFAR10 we reduce the ResNet32 model by 75% with no loss in quality, and are able to do a 10x compression while still achieving above 90% accuracy.