Unsupervised Feature Selection based on Adaptive Similarity Learning and Subspace Clustering Mohsen Ghassemi Parsa a, Hadi Zare a,, Mehdi Ghatee b a Faculty of New Sciences and Technologies, University of Tehran, Iran b Department of Computer Science, Amirkabir University of Technology, IranAbstract Feature selection methods have an important role on the readability of data and the reduction of complexity of learning algorithms. In recent years, a variety of efforts are investigated on feature selection problems based on unsupervised viewpoint due to the laborious labeling task on large datasets. In this paper, we propose a novel approach on unsupervised feature selection initiated from the subspace clustering to preserve the similarities by representation learning of low dimensional subspaces among the samples. A self-expressive model is employed to implicitly learn the cluster similarities in an adaptive manner. The proposed method not only maintains the sample similarities through subspace clustering, but it also captures the discriminative information based on a regularized regression model. In line with the convergence analysis of the proposed method, the experimental results on benchmark datasets demonstrate the effectiveness of our approach as compared with the state of the art methods.

Recent advancements in language representation models such as BERT have led to a rapid improvement in numerous natural language processing tasks. However, language models usually consist of a few hundred million trainable parameters with embedding space distributed across multiple layers, thus making them challenging to be fine-tuned for a specific task or to be transferred to a new domain. To determine whether there are task-specific neurons that can be exploited for unsupervised transfer learning, we introduce a method for selecting the most important neurons to solve a specific classification task. This algorithm is further extended to multi-source transfer learning by computing the importance of neurons for several single-source transfer learning scenarios between different subsets of data sources. Besides, a task-specific fingerprint for each data source is obtained based on the percentage of the selected neurons in each layer. We perform extensive experiments in unsupervised transfer learning for sentiment analysis, natural language inference and sentence similarity, and compare our results with the existing literature and baselines. Significantly, we found that the source and target data sources with higher degrees of similarity between their task-specific fingerprints demonstrate a better transferability property. We conclude that our method can lead to better performance using just a few hundred task-specific and interpretable neurons.

We study the problem of off-policy critic evaluation in several variants of value-based off-policy actor-critic algorithms. Off-policy actor-critic algorithms require an off-policy critic evaluation step, to estimate the value of the new policy after every policy gradient update. Despite enormous success of off-policy policy gradients on control tasks, existing general methods suffer from high variance and instability, partly because the policy improvement depends on gradient of the estimated value function. In this work, we present a new way of off-policy policy evaluation in actor-critic, based on the doubly robust estimators. We extend the doubly robust estimator from off-policy policy evaluation (OPE) to actor-critic algorithms that consist of a reward estimator performance model. We find that doubly robust estimation of the critic can significantly improve performance in continuous control tasks. Furthermore, in cases where the reward function is stochastic that can lead to high variance, doubly robust critic estimation can improve performance under corrupted, stochastic reward signals, indicating its usefulness for robust and safe reinforcement learning.

Earth Signals and Systems Group, Department of Earth, Atmospheric, and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA 91106, USA (Dated: December 12, 2019) In modeling nonlinear dynamics, neural networks are of interest for prediction and uncertainty quantification. The "learnability" of chaotic dynamics by neural networks, however, remains poorly understood. In this work, we show that a parsimonious network trained on few data points suffices for accurate prediction of local divergence rates on the whole attractor. To understand neural learnability, we decompose the mappings in the neural network into a series of geometric stretching and compressing operations that indicate topological mixing and, therefore, chaos. This reveals that neural networks and chaotic dynamical systems are structurally similar, which yields excellent reproduction of local divergence rates. To build parsimonious networks, we employ an approach that matches the spectral features of the dynamics of deep learning those of polynomial regression.

Deep Neural Networks have achieved remarkable success relying on the developing high computation capability of GPUs and large-scale datasets with increasing network depth and width in image recognition, object detection and many other applications. However, due to the expensive computation and intensive memory, researchers have concentrated on designing compression methods in recent years. In this paper, we briefly summarize the existing advanced techniques that are useful in model compression at first. After that, we give a detailed description on group lasso regularization and its variants. More importantly, we propose an improving framework of partial regularization based on the relationship between neurons and connections of adjacent layers. It is reasonable and feasible with the help of permutation property of neural network . Experiment results show that partial regularization methods brings improvements such as higher classification accuracy in both training and testing stages on multiple datasets. Since our regularizers contain the computation of less parameters, it shows competitive performances in terms of the total running time of experiments. Finally, we analysed the results and draw a conclusion that the optimal network structure must exist and depend on the input data.

As deep neural networks become more adept at traditional tasks, many of the most exciting new challenges concern multimodality--observations that combine diverse types, such as image and text. In this paper, we introduce a family of multimodal deep generative models derived from variational bounds on the evidence (data marginal likelihood). As part of our derivation we find that many previous multimodal variational autoencoders used objectives that do not correctly bound the joint marginal likelihood across modalities. We further generalize our objective to work with several types of deep generative model (V AE, GAN, and flow-based), and allow use of different model types for different modalities. We benchmark our models across many image, label, and text datasets, and find that our multimodal V AEs excel with and without weak supervision. Additional improvements come from use of GAN image models with V AE language models. Finally, we investigate the effect of language on learned image representations through a variety of downstream tasks, such as compositionally, bounding box prediction, and visual relation prediction. We find evidence that these image representations are more abstract and compositional than equivalent representations learned from only visual data.

--Sentiment Analysis of microblog feeds has attracted considerable interest in recent times. Most of the current work focuses on tweet sentiment classification. But not much work has been done to explore how reliable the opinions of the mass (crowd wisdom) in social network microblogs such as twitter are in predicting outcomes of certain events such as election debates. In this work, we investigate whether crowd wisdom is useful in predicting such outcomes and whether their opinions are influenced by the experts in the field. We work in the domain of multi-label classification to perform sentiment classification of tweets and obtain the opinion of the crowd. This learnt sentiment is then used to predict outcomes of events such as: US Presidential Debate winners, Grammy A ward winners, Super Bowl Winners. We find that in most of the cases, the wisdom of the crowd does indeed match with that of the experts, and in cases where they don't (particularly in the case of debates), we see that the crowd's opinion is actually influenced by that of the experts. I NTRODUCTION Over the past few years, microblogs have become one of the most popular online social networks. Microblogging websites have evolved to become a source of varied kinds of information. This is due to the nature of microblogs: people post real-time messages about their opinions and express sentiment on a variety of topics, discuss current issues, complain, etc. Twitter is one such popular microblogging service where users create status messages (called "tweets"). With over 400 million tweets per day on Twitter, microblog users generate large amount of data, which cover very rich topics ranging from politics, sports to celebrity gossip. Because the user generated content on microblogs covers rich topics and expresses sentiment/opinions of the mass, mining and analyzing this information can prove to be very beneficial both to the industrial and the academic community. Tweet classification has attracted considerable attention because it has become very important to analyze peoples' sentiments and opinions over social networks.

Allowing members of the crowd to propose novel microtasks for one another is an effective way to combine the efficiencies of traditional microtask work with the inventiveness and hypothesis generation potential of human workers. However, microtask proposal leads to a growing set of tasks that may overwhelm limited crowdsourcer resources. Crowdsourcers can employ methods to utilize their resources efficiently, but algorithmic approaches to efficient crowdsourcing generally require a fixed task set of known size. In this paper, we introduce *cost forecasting* as a means for a crowdsourcer to use efficient crowdsourcing algorithms with a growing set of microtasks. Cost forecasting allows the crowdsourcer to decide between eliciting new tasks from the crowd or receiving responses to existing tasks based on whether or not new tasks will cost less to complete than existing tasks, efficiently balancing resources as crowdsourcing occurs. Experiments with real and synthetic crowdsourcing data show that cost forecasting leads to improved accuracy. Accuracy and efficiency gains for crowd-generated microtasks hold the promise to further leverage the creativity and wisdom of the crowd, with applications such as generating more informative and diverse training data for machine learning applications and improving the performance of user-generated content and question-answering platforms.

Ordinal regression falls between discrete-valued classification and continuous-valued regression. Ordinal target variables can be associated with ranked random variables. These random variables are known as order statistics and they are closely related to ordinal regression. However, the challenge of using order statistics for ordinal regression prediction is finding a suitable parent distribution. In this work, we provide a case study of a real-world orienteering relay race by viewing it as a random process. For this process, we show that accurate order statistical ordinal regression predictions of final team rankings, or places, can be obtained by assuming a lognormal distribution of individual leg times. Moreover, we apply Fenton-Wilkinson approximations to intermediate changeover times alongside an estimator for the total number of teams as in the notorious German tank problem. The purpose of this work is, in part, to spark interest in studying the applicability of order statistics in ordinal regression problems.

A BSTRACT When performing imitation learning from expert demonstrations, distribution matching is a popular approach, in which one alternates between estimating distribution ratios and then using these ratios as rewards in a standard reinforcement learning (RL) algorithm. Traditionally, estimation of the distribution ratio requires on-policy data, which has caused previous work to either be exorbitantly data-inefficient or alter the original objective in a manner that can drastically change its optimum. In this work, we show how the original distribution ratio estimation objective may be transformed in a principled manner to yield a completely off-policy objective. In addition to the data-efficiency that this provides, we are able to show that this objective also renders the use of a separate RL optimization unnecessary. Rather, an imitation policy may be learned directly from this objective without the use of explicit rewards. We call the resulting algorithm V alueDICEand evaluate it on a suite of popular imitation learning benchmarks, finding that it can achieve state-of-the-art sample efficiency and performance. Accordingly, many successful demonstrations of RL often rely on carefully handcrafted rewards with various bonuses and penalties designed to encourage intended behavior (Nachum et al., 2019a; Andrychowicz et al., 2018).