Efficient training of lightweight neural networks using Online Self-Acquired Knowledge Distillation

Knowledge Distillation has been established as a highly promising approach for training compact and faster models by transferring knowledge from heavyweight and powerful models. However, KD in its conventional version constitutes an enduring, computationally and memory demanding process. In this paper, Online Self-Acquired Knowledge Distillation (OSAKD) is proposed, aiming to improve the performance of any deep neural model in an online manner. We utilize k-nn non-parametric density estimation technique for estimating the unknown probability distributions of the data samples in the output feature space. This allows us for directly estimating the posterior class probabilities of the data samples, and we use them as soft labels that encode explicit information about the similarities of the data with the classes, negligibly affecting the computational cost. The experimental evaluation on four datasets validates the effectiveness of proposed method.

Quadratic mutual information regularization in real-time deep CNN models

In this paper, regularized lightweight deep convolutional neural network models, capable of effectively operating in real-time on devices with restricted computational power for high-resolution video input are proposed. Furthermore, a novel regularization method motivated by the Quadratic Mutual Information, in order to improve the generalization ability of the utilized models is proposed. Extensive experiments on various binary classification problems involved in autonomous systems are performed, indicating the effectiveness of the proposed models as well as of the proposed regularizer.

Semantic Scene Segmentation for Robotics Applications

Semantic scene segmentation plays a critical role in a wide range of robotics applications, e.g., autonomous navigation. These applications are accompanied by specific computational restrictions, e.g., operation on low-power GPUs, at sufficient speed, and also for high-resolution input. Existing state-of-the-art segmentation models provide evaluation results under different setups and mainly considering high-power GPUs. In this paper, we investigate the behavior of the most successful semantic scene segmentation models, in terms of deployment (inference) speed, under various setups (GPUs, input sizes, etc.) in the context of robotics applications. The target of this work is to provide a comparative study of current state-of-the-art segmentation models so as to select the most compliant with the robotics applications requirements.

Attention-based Neural Bag-of-Features Learning for Sequence Data

In this paper, we propose 2D-Attention (2DA), a generic attention formulation for sequence data, which acts as a complementary computation block that can detect and focus on relevant sources of information for the given learning objective. The proposed attention module is incorporated into the recently proposed Neural Bag of Feature (NBoF) model to enhance its learning capacity. Since 2DA acts as a plug-in layer, injecting it into different computation stages of the NBoF model results in different 2DA-NBoF architectures, each of which possesses a unique interpretation. We conducted extensive experiments in financial forecasting, audio analysis as well as medical diagnosis problems to benchmark the proposed formulations in comparison with existing methods, including the widely used Gated Recurrent Units. Our empirical analysis shows that the proposed attention formulations can not only improve performances of NBoF models but also make them resilient to noisy data.

Heterogeneous Knowledge Distillation using Information Flow Modeling

Knowledge Distillation (KD) methods are capable of transferring the knowledge encoded in a large and complex teacher into a smaller and faster student. Early methods were usually limited to transferring the knowledge only between the last layers of the networks, while latter approaches were capable of performing multi-layer KD, further increasing the accuracy of the student. However, despite their improved performance, these methods still suffer from several limitations that restrict both their efficiency and flexibility. First, existing KD methods typically ignore that neural networks undergo through different learning phases during the training process, which often requires different types of supervision for each one. Furthermore, existing multi-layer KD methods are usually unable to effectively handle networks with significantly different architectures (heterogeneous KD). In this paper we propose a novel KD method that works by modeling the information flow through the various layers of the teacher model and then train a student model to mimic this information flow. The proposed method is capable of overcoming the aforementioned limitations by using an appropriate supervision scheme during the different phases of the training process, as well as by designing and training an appropriate auxiliary teacher model that acts as a proxy model capable of explaining the way the teacher works to the student. The effectiveness of the proposed method is demonstrated using four image datasets and several different evaluation setups.

Deep Adaptive Input Normalization for Time Series Forecasting

Deep Learning (DL) models can be used to tackle time series analysis tasks with great success. However, the performance of DL models can degenerate rapidly if the data are not appropriately normalized. This issue is even more apparent when DL is used for financial time series forecasting tasks, where the non-stationary and multimodal nature of the data pose significant challenges and severely affect the performance of DL models. In this work, a simple, yet effective, neural layer, that is capable of adaptively normalizing the input time series, while taking into account the distribution of the data, is proposed. The proposed layer is trained in an end-to-end fashion using back-propagation and leads to significant performance improvements compared to other evaluated normalization schemes. The proposed method differs from traditional normalization methods since it learns how to perform normalization for a given task instead of using a fixed normalization scheme. At the same time, it can be directly applied to any new time series without requiring re-training. The effectiveness of the proposed method is demonstrated using a large-scale limit order book dataset, as well as a load forecasting dataset.

Temporal Logistic Neural Bag-of-Features for Financial Time series Forecasting leveraging Limit Order Book Data

Time series forecasting is a crucial component of many important applications, ranging from forecasting the stock markets to energy load prediction. The high-dimensionality, velocity and variety of the data collected in these applications pose significant and unique challenges that must be carefully addressed for each of them. In this work, a novel Temporal Logistic Neural Bag-of-Features approach, that can be used to tackle these challenges, is proposed. The proposed method can be effectively combined with deep neural networks, leading to powerful deep learning models for time series analysis. However, combining existing BoF formulations with deep feature extractors pose significant challenges: the distribution of the input features is not stationary, tuning the hyper-parameters of the model can be especially difficult and the normalizations involved in the BoF model can cause significant instabilities during the training process. The proposed method is capable of overcoming these limitations by a employing a novel adaptive scaling mechanism and replacing the classical Gaussian-based density estimation involved in the regular BoF model with a logistic kernel. The effectiveness of the proposed approach is demonstrated using extensive experiments on a large-scale financial time series dataset that consists of more than 4 million limit orders.

Deep Supervised Hashing leveraging Quadratic Spherical Mutual Information for Content-based Image Retrieval

Several deep supervised hashing techniques have been proposed to allow for efficiently querying large image databases. However, deep supervised image hashing techniques are developed, to a great extent, heuristically often leading to suboptimal results. Contrary to this, we propose an efficient deep supervised hashing algorithm that optimizes the learned codes using an information-theoretic measure, the Quadratic Mutual Information (QMI). The proposed method is adapted to the needs of large-scale hashing and information retrieval leading to a novel information-theoretic measure, the Quadratic Spherical Mutual Information (QSMI). Apart from demonstrating the effectiveness of the proposed method under different scenarios and outperforming existing state-of-the-art image hashing techniques, this paper provides a structured way to model the process of information retrieval and develop novel methods adapted to the needs of each application.

Style Decomposition for Improved Neural Style Transfer

Universal Neural Style Transfer (NST) methods are capable of performing style transfer of arbitrary styles in a style-agnostic manner via feature transforms in (almost) real-time. Even though their unimodal parametric style modeling approach has been proven adequate to transfer a single style from relatively simple images, they are usually not capable of effectively handling more complex styles, producing significant artifacts, as well as reducing the quality of the synthesized textures in the stylized image. To overcome these limitations, in this paper we propose a novel universal NST approach that separately models each sub-style that exists in a given style image (or a collection of style images). This allows for better modeling the subtle style differences within the same style image and then using the most appropriate sub-style (or mixtures of different sub-styles) to stylize the content image. The ability of the proposed approach to a) perform a wide range of different stylizations using the sub-styles that exist in one style image, while giving the ability to the user to appropriate mix the different sub-styles, b) automatically match the most appropriate sub-style to different semantic regions of the content image, improving existing state-of-the-art universal NST approaches, and c) detecting and transferring the sub-styles from collections of images are demonstrated through extensive experiments.

Decoding Generic Visual Representations From Human Brain Activity using Machine Learning

Among the most impressive recent applications of neural decoding is the visual representation decoding, where the category of an object that a subject either sees or imagines is inferred by observing his/her brain activity. Even though there is an increasing interest in the aforementioned visual representation decoding task, there is no extensive study of the effect of using different machine learning models on the decoding accuracy. In this paper we provide an extensive evaluation of several machine learning models, along with different similarity metrics, for the aforementioned task, drawing many interesting conclusions. That way, this paper a) paves the way for developing more advanced and accurate methods and b) provides an extensive and easily reproducible baseline for the aforementioned decoding task.