A Berkeley View of Systems Challenges for AI

With the increasing commoditization of computer vision, speech recognition and machine translation systems and the widespread deployment of learning-based back-end technologies such as digital advertising and intelligent infrastructures, AI (Artificial Intelligence) has moved from research labs to production. These changes have been made possible by unprecedented levels of data and computation, by methodological advances in machine learning, by innovations in systems software and architectures, and by the broad accessibility of these technologies. The next generation of AI systems promises to accelerate these developments and increasingly impact our lives via frequent interactions and making (often mission-critical) decisions on our behalf, often in highly personalized contexts. Realizing this promise, however, raises daunting challenges. In particular, we need AI systems that make timely and safe decisions in unpredictable environments, that are robust against sophisticated adversaries, and that can process ever increasing amounts of data across organizations and individuals without compromising confidentiality. These challenges will be exacerbated by the end of the Moore's Law, which will constrain the amount of data these technologies can store and process. In this paper, we propose several open research directions in systems, architectures, and security that can address these challenges and help unlock AI's potential to improve lives and society.

Disentangling Dynamics and Content for Control and Planning

In this paper, We study the problem of learning a controllable representation for high-dimensional observations of dynamical systems. Specifically, we consider a situation where there are multiple sets of observations of dynamical systems with identical underlying dynamics. Only one of these sets has information about the effect of actions on the observation and the rest are just some random observations of the system. Our goal is to utilize the information in that one set and find a representation for the other sets that can be used for planning and ling-term prediction.

JADE: Joint Autoencoders for Dis-Entanglement

The problem of feature disentanglement has been explored in the literature, for the purpose of image and video processing and text analysis. State-of-the-art methods for disentangling feature representations rely on the presence of many labeled samples. In this work, we present a novel method for disentangling factors of variation in data-scarce regimes. Specifically, we explore the application of feature disentangling for the problem of supervised classification in a setting where few labeled samples exist, and there are no unlabeled samples for use in unsupervised training. Instead, a similar datasets exists which shares at least one direction of variation with the sample-constrained datasets. We train our model end-to-end using the framework of variational autoencoders and are able to experimentally demonstrate that using an auxiliary dataset with similar variation factors contribute positively to classification performance, yielding competitive results with the state-of-the-art in unsupervised learning.

Improving the Accuracy of Pre-trained Word Embeddings for Sentiment Analysis

Sentiment analysis is one of the well-known tasks and fast growing research areas in natural language processing (NLP) and text classifications. This technique has become an essential part of a wide range of applications including politics, business, advertising and marketing. There are various techniques for sentiment analysis, but recently word embeddings methods have been widely used in sentiment classification tasks. Word2Vec and GloVe are currently among the most accurate and usable word embedding methods which can convert words into meaningful vectors. However, these methods ignore sentiment information of texts and need a huge corpus of texts for training and generating exact vectors which are used as inputs of deep learning models. As a result, because of the small size of some corpuses, researcher often have to use pre-trained word embeddings which were trained on other large text corpus such as Google News with about 100 billion words. The increasing accuracy of pre-trained word embeddings has a great impact on sentiment analysis research. In this paper we propose a novel method, Improved Word Vectors (IWV), which increases the accuracy of pre-trained word embeddings in sentiment analysis. Our method is based on Part-of-Speech (POS) tagging techniques, lexicon-based approaches and Word2Vec/GloVe methods. We tested the accuracy of our method via different deep learning models and sentiment datasets. Our experiment results show that Improved Word Vectors (IWV) are very effective for sentiment analysis.

Deep Structure for end-to-end inverse rendering

Inverse rendering in a 3D format denoted to recovering the 3D properties of a scene given 2D input image(s) and is typically done using 3D Morphable Model (3DMM) based methods from single view images. These models formulate each face as a weighted combination of some basis vectors extracted from the training data. In this paper a deep framework is proposed in which the coefficients and basis vectors are computed by training an autoencoder network and a Convolutional Neural Network (CNN) simultaneously. The idea is to find a common cause which can be mapped to both the 3D structure and corresponding 2D image using deep networks. The empirical results verify the power of deep framework in finding accurate 3D shapes of human faces from their corresponding 2D images on synthetic datasets of human faces.

Fast Spectral Clustering Using Autoencoders and Landmarks

In this paper, we introduce an algorithm for performing spectral clustering efficiently. Spectral clustering is a powerful clustering algorithm that suffers from high computational complexity, due to eigen decomposition. In this work, we first build the adjacency matrix of the corresponding graph of the dataset. To build this matrix, we only consider a limited number of points, called landmarks, and compute the similarity of all data points with the landmarks. Then, we present a definition of the Laplacian matrix of the graph that enable us to perform eigen decomposition efficiently, using a deep autoencoder. The overall complexity of the algorithm for eigen decomposition is $O(np)$, where $n$ is the number of data points and $p$ is the number of landmarks. At last, we evaluate the performance of the algorithm in different experiments.

Semi-Supervised Representation Learning based on Probabilistic Labeling

In this paper, we present a new algorithm for semi-supervised representation learning. In this algorithm, we first find a vector representation for the labels of the data points based on their local positions in the space. Then, we map the data to lower-dimensional space using a linear transformation such that the dependency between the transformed data and the assigned labels is maximized. In fact, we try to find a mapping that is as discriminative as possible. The approach will use Hilber-Schmidt Independence Criterion (HSIC) as the dependence measure. We also present a kernelized version of the algorithm, which allows non-linear transformations and provides more flexibility in finding the appropriate mapping. Use of unlabeled data for learning new representation is not always beneficial and there is no algorithm that can deterministically guarantee the improvement of the performance by exploiting unlabeled data. Therefore, we also propose a bound on the performance of the algorithm, which can be used to determine the effectiveness of using the unlabeled data in the algorithm. We demonstrate the ability of the algorithm in finding the transformation using both toy examples and real-world datasets.

Semi-supervised Dictionary Learning Based on Hilbert-Schmidt Independence Criterion

In this paper, a novel semi-supervised dictionary learning and sparse representation (SS-DLSR) is proposed. The proposed method benefits from the supervisory information by learning the dictionary in a space where the dependency between the data and class labels is maximized. This maximization is performed using Hilbert-Schmidt independence criterion (HSIC). On the other hand, the global distribution of the underlying manifolds were learned from the unlabeled data by minimizing the distances between the unlabeled data and the corresponding nearest labeled data in the space of the dictionary learned. The proposed SS-DLSR algorithm has closed-form solutions for both the dictionary and sparse coefficients, and therefore does not have to learn the two iteratively and alternately as is common in the literature of the DLSR. This makes the solution for the proposed algorithm very fast. The experiments confirm the improvement in classification performance on benchmark datasets by including the information from both labeled and unlabeled data, particularly when there are many unlabeled data.

On the Invariance of Dictionary Learning and Sparse Representation to Projecting Data to a Discriminative Space

In this paper, it is proved that dictionary learning and sparse representation is invariant to a linear transformation. It subsumes the special case of transforming/projecting the data into a discriminative space. This is important because recently, supervised dictionary learning algorithms have been proposed, which suggest to include the category information into the learning of dictionary to improve its discriminative power. Among them, there are some approaches that propose to learn the dictionary in a discriminative projected space. To this end, two approaches have been proposed: first, assigning the discriminative basis as the dictionary and second, perform dictionary learning in the projected space. Based on the invariance of dictionary learning to any transformation in general, and to a discriminative space in particular, we advocate the first approach.

Supervised Dictionary Learning and Sparse Representation-A Review

Dictionary learning and sparse representation (DLSR) is a recent and successful mathematical model for data representation that achieves state-of-the-art performance in various fields such as pattern recognition, machine learning, computer vision, and medical imaging. The original formulation for DLSR is based on the minimization of the reconstruction error between the original signal and its sparse representation in the space of the learned dictionary. Although this formulation is optimal for solving problems such as denoising, inpainting, and coding, it may not lead to optimal solution in classification tasks, where the ultimate goal is to make the learned dictionary and corresponding sparse representation as discriminative as possible. This motivated the emergence of a new category of techniques, which is appropriately called supervised dictionary learning and sparse representation (S-DLSR), leading to more optimal dictionary and sparse representation in classification tasks. Despite many research efforts for S-DLSR, the literature lacks a comprehensive view of these techniques, their connections, advantages and shortcomings. In this paper, we address this gap and provide a review of the recently proposed algorithms for S-DLSR. We first present a taxonomy of these algorithms into six categories based on the approach taken to include label information into the learning of the dictionary and/or sparse representation. For each category, we draw connections between the algorithms in this category and present a unified framework for them. We then provide guidelines for applied researchers on how to represent and learn the building blocks of an S-DLSR solution based on the problem at hand. This review provides a broad, yet deep, view of the state-of-the-art methods for S-DLSR and allows for the advancement of research and development in this emerging area of research.