Search

Stochastic multi-armed bandit (MAB) mechanisms are widely used in sponsored search auctions, crowdsourcing, online procurement, etc. Existing stochastic MAB mechanisms with a deterministic payment rule, proposed in the literature, necessarily suffer a regret of Ω(T2/3), where T is the number of time steps. This happens because the existing mechanisms consider the worst case scenario where the means of the agents’ stochastic rewards are separated by a very small amount that depends on T. We make, and, exploit the crucial observation that in most scenarios, the separation between the agents’ rewards is rarely a function of T. Moreover, in the case that the rewards of the arms are arbitrarily close, the regret contributed by such sub-optimal arms is minimal. Our idea is to allow the center to indicate the resolution, Δ, with which the agents must be distinguished. This immediately leads us to introduce the notion of Δ-Regret. Using sponsored search auctions as a concrete example (the same idea applies for other applications as well), we propose a dominant strategy incentive compatible (DSIC) and individually rational (IR), deterministic MAB mechanism, based on ideas from the Upper Confidence Bound (UCB) family of MAB algorithms. Remarkably, the proposed mechanism Δ-UCB achieves a Δ-regret of $O(\log T)$ for the case of sponsored search auctions. We first establish the results for single slot sponsored search auctions and then non-trivially extend the results to the case where multiple slots are to be allocated.

Checkpoints for the model described in the paper - Efficient Constituency Tree based Encoding for Natural Language to Bash Translation (to be published in NAACL 2022) The checkpoints are named by the data split they were trained on.

Research on the task of Reading Comprehension style Question Answering (RCQA) has gained momentum in recent years due to the emergence of human annotated datasets and associated leaderboards, for example CoQA, HotpotQA, SQuAD, TriviaQA, etc. While state-of-the-art has advanced considerably, there is still ample opportunity to advance it further on some important variants of the RCQA task. In this paper, we propose a novel deep neural architecture, called TAP (Translucent Answer Prediction), to identify answers and evidence (in the form of supporting facts) in an RCQA task requiring multi-hop reasoning. TAP comprises two loosely coupled networks – Local and Global Interaction eXtractor (LoGIX) and Answer Predictor (AP). LoGIX predicts supporting facts, whereas AP consumes these predicted supporting facts to predict the answer span. The novel design of LoGIX is inspired by two key design desiderata – local context and global interaction– that we identified by analyzing examples of multi-hop RCQA task. The loose coupling between LoGIX and the AP reveals the set of sentences used by the AP in predicting an answer. Therefore, answer predictions of TAP can be interpreted in a translucent manner. TAP offers state-of-the-art performance on the HotpotQA (Yang et al. 2018) dataset – an apt dataset for multi-hop RCQA task – as it occupies Rank-1 on its leaderboard (https://hotpotqa.github.io/) at the time of submission.

Machine learning models are increasingly being deployed in practice. Machine Learning as a Service (MLaaS) providers expose such models to queries by third-party developers through application programming interfaces (APIs). Prior work has developed model extraction attacks, in which an attacker extracts an approximation of an MLaaS model by making black-box queries to it. We design ActiveThief – a model extraction framework for deep neural networks that makes use of active learning techniques and unannotated public datasets to perform model extraction. It does not expect strong domain knowledge or access to annotated data on the part of the attacker. We demonstrate that (1) it is possible to use ActiveThief to extract deep classifiers trained on a variety of datasets from image and text domains, while querying the model with as few as 10-30% of samples from public datasets, (2) the resulting model exhibits a higher transferability success rate of adversarial examples than prior work, and (3) the attack evades detection by the state-of-the-art model extraction detection method, PRADA.

Explainable Recommendations provide the reasons behind why an item is recommended to a user, which often leads to increased user satisfaction and persuasiveness. An intuitive way to explain recommendations is by generating a synthetic personalized natural language review for a user-item pair. Although there exist some approaches in the literature that explain recommendations by generating reviews, the quality of the reviews is questionable. Besides, these methods usually take considerable time to train the underlying language model responsible for generating the text. In this work, we propose ReXPlug, an end-to-end framework with a plug and play way of explaining recommendations. ReXPlug predicts accurate ratings as well as exploits Plug and Play Language Model to generate high-quality reviews. We train a simple sentiment classifier for controlling a pre-trained language model for the generation, bypassing the language model's training from scratch again. Such a simple and neat model is much easier to implement and train, and hence, very efficient for generating reviews. We personalize the reviews by leveraging a special jointly-trained cross attention network. Our detailed experiments show that ReXPlug outperforms many recent models across various datasets on rating prediction by utilizing textual reviews as a regularizer. Quantitative analysis shows that the reviews generated by ReXPlug are semantically close to the ground truth reviews, while the qualitative analysis demonstrates the high quality of the generated reviews, both from empirical and analytical viewpoints. Our implementation is available online.

Novice programmers often struggle with the formal syntax of programming languages. In the traditional classroom setting, they can make progress with the help of real time feedback from their instructors which is often impossible to get in the massive open online course (MOOC) setting. Syntactic error repair techniques have huge potential to assist them at scale. Towards this, we design a novel programming language correction framework amenable to reinforcement learning. The framework allows an agent to mimic human actions for text navigation and editing. We demonstrate that the agent can be trained through self-exploration directly from the raw input, that is, program text itself, without either supervision or any prior knowledge of the formal syntax of the programming language. We evaluate our technique on a publicly available dataset containing 6975 erroneous C programs with typographic errors, written by students during an introductory programming course. Our technique fixes 1699 (24.4%) programs completely and 1310 (18.8%) program partially, outperforming DeepFix, a state-of-the-art syntactic error repair technique, which uses a fully supervised neural machine translation approach.

Cross-Domain Collaborative Filtering (CDCF) provides a way to alleviate data sparsity and cold-start problems present in recommendation systems by exploiting the knowledge from related domains. Existing CDCF models are either based on matrix factorization or deep neural networks. Independent use of either of the techniques in isolation may result in suboptimal performance for the prediction task. Also, most of the existing models face challenges particularly in handling diversity between domains and learning complex non-linear relationships that exist amongst entities (users/items) within and across domains. In this work, we propose an end-to-end neural network model – NeuCDCF, to address these challenges in a cross-domain setting. More importantly, NeuCDCF is based on a wide and deep framework and learns the representations jointly using both matrix factorization and deep neural networks. We perform experiments on four real-world datasets and demonstrate that our model performs better than state-of-the-art CDCF models.

Bundle recommendation – recommending a group of products in place of individual products to customers is gaining attention day by day. It presents two interesting challenges – (1) how to personalize and recommend existing bundles to users, and (2) how to generate personalized novel bundles targeting specific users. Recently, few models have been proposed for modeling the bundle recommendation problem. However, they have the following shortcomings. First, they do not consider the higher-order relationships amongst the entities (users, items and bundles). Second, they do not model the relative influence of items present in the bundles, which is crucial in defining such bundles. In this work, we propose GRAM-SMOT – a graph attention-based framework to address the above challenges. Further, we define a loss function based on the metric-learning approach to learn the embeddings of entities efficiently. To generate novel bundles, we propose a strategy that leverages submodular function maximization. To analyze the performance of the proposed model, we conduct comprehensive experiments on two real-world datasets. The experimental results demonstrate the superior performance of the proposed model over the existing state-of-the-art models.

Exploiting heterogeneous information networks (HIN) to top-N recommendation has been shown to alleviate the data sparsity problem present in recommendation systems. This requires careful effort in extracting relevant knowledge from HIN. However, existing models in this setting have the following shortcomings. Mainly, they are not end-to-end, which puts the burden on the system to first learn similarity or commuting matrix offline using some manually selected meta-paths before we train for the top-N recommendation objective. Further, they do not attentively extract user-specific information from HIN, which is essential for personalization. To address these challenges, we propose an end-to-end neural network model – GAMMA (Graph and Multi-view Memory Attention mechanism). We aim to replace the offline meta-path based similarity or commuting matrix computation with a graph attention mechanism. Besides, with different semantics of items in HIN, we propose a multi-view memory attention mechanism to learn more profound user-specific item views. Experiments on three real-world datasets demonstrate the effectiveness of our model for top-N recommendation setting.

Social recommendation systems typically combine extra information like a social network with the user-item interaction network in order to alleviate data sparsity issues. This also helps in making more accurate and personalized recommendations. However, most of the existing systems work under the assumption that all socially connected users have equal influence on each other in a social network, which is not true in practice. Further, estimating the quantum of influence that exists among entities in a user-item interaction network is essential when only implicit ratings are available. This has been ignored even in many recent state-of-the-art models such as SAMN (Social Attentional Memory Network) and DeepSoR (Deep neural network model on Social Relations). Many a time, capturing a complex relationship between the entities (users/items) is essential to boost the performance of a recommendation system. We address these limitations by proposing a novel neural network model, SoRecGAT, which employs multi-head and multi-layer graph attention mechanism. The attention mechanism helps the model learn the influence of entities on each other more accurately. The proposed model also takes care of heterogeneity among the entities seamlessly. SoRecGAT is a general approach and we also validate its suitability when information in the form of a network of co-purchased items is available. Empirical results on eight real-world datasets demonstrate that the proposed model outperforms state-of-the-art models.

Cross-Domain Collaborative Filtering (CDCF) mitigates data sparsity and cold-start issues present in conventional recommendation systems by exploiting and transferring knowledge from related domains. Leveraging user-generated tags (e.g. ancient-literature, military-history) for bridging the related domains is becoming a popular way for enhancing personalized recommendations. However, existing tag based models bridge the domains based on common tags between domains and their co-occurrence frequencies. This results in capturing the syntax similarities between the tags and ignoring the semantic similarities between them. In this work, to address these, we propose TagEmbedSVD, a tag-based CDCF model to cross-domain setting. TagEmbedSVD makes use of the pre-trained word embeddings (word2vec) for tags to enhance personalized recommendations in the cross-domain setting. Empirical evaluation on two real-world datasets demonstrates that our proposed model performs better than the existing tag based CDCF models.

In this work, we propose an automated method to identify semantic bugs in student programs, called ATAS, which builds upon the recent advances in both symbolic execution and active learning. Symbolic execution is a program analysis technique which can generate test cases through symbolic constraint solving. Our method makes use of a reference implementation of the task as its sole input. We compare our method with a symbolic execution-based baseline on 6 programming tasks retrieved from CodeForces comprising a total of 23 K student submissions. We show an average improvement of over 2.5x over the baseline in terms of runtime (thus making it more suitable for online evaluation), without a significant degradation in evaluation accuracy.

Multi-label classification has attracted much interest due to its wide applicability. Modeling label interactions and investigating their impact on classifier quality are crucial aspects of multi-label classification. In this paper, we propose a multi-structure SVM (called MSSVM) which allows the user to hypothesize multiple label interaction structures and helps to identify their importance in improving generalization performance. We design an efficient optimization algorithm to solve the proposed MSSVM. Extensive empirical evaluation provides fresh and interesting insights into the following questions: (a) How do label interactions affect multiple performance metrics typically used in multi-label classification? (b) Do higher order label interactions significantly impact a given performance metric for a particular dataset? (c) Can we make useful suggestions on the label interaction structure? and (d) Is it always beneficial to model label interactions in multi-label classification?

In structured output learning, obtaining labeled data for real-world applications is usually costly, while unlabeled examples are available in abundance. Semisupervised structured classification deals with a small number of labeled examples and a large number of unlabeled structured data. In this work, we consider semisupervised structural support vector machines with domain constraints. The optimization problem, which in general is not convex, contains the loss terms associated with the labeled and unlabeled examples, along with the domain constraints. We propose a simple optimization approach that alternates between solving a supervised learning problem and a constraint matching problem. Solving the constraint matching problem is difficult for structured prediction, and we propose an efficient and effective label switching method to solve it. The alternating optimization is carried out within a deterministic annealing framework, which helps in effective constraint matching and avoiding poor local minima, which are not very useful. The algorithm is simple and easy to implement. Further, it is suitable for any structured output learning problem where exact inference is available. Experiments on benchmark sequence labeling data sets and a natural language parsing data set show that the proposed approach, though simple, achieves comparable generalization performance.

Community-driven Question Answering (CQA) systems such as Yahoo! Answers have become valuable sources of reusable information. CQA retrieval enables usage of historical CQA archives to solve new questions posed by users. This task has received much recent attention, with methods building upon literature from translation models, topic models, and deep learning. In this paper, we devise a CQA retrieval technique, LASER-QA, that embeds question-answer pairs within a unified latent space preserving the local neighborhood structure of question and answer spaces. The idea is that such a space mirrors semantic similarity among questions as well as answers, thereby enabling high quality retrieval. Through an empirical analysis on various real-world QA datasets, we illustrate the improved effectiveness of LASER-QA over state-of-the-art methods.

Multi-label classification is a well-known supervised machine learning setting where each instance is associated with multiple classes. Examples include annotation of images with multiple labels, assigning multiple tags for a web page, etc. Since several labels can be assigned to a single instance, one of the key challenges in this problem is to learn the correlations between the classes. Our first contribution assumes labels from a perfect source. Towards this, we propose a novel topic model (ML-PA-LDA). The distinguishing feature in our model is that classes that are present as well as the classes that are absent generate the latent topics and hence the words. Extensive experimentation on real world datasets reveals the superior performance of the proposed model. A natural source for procuring the training dataset is through mining user-generated content or directly through users in a crowdsourcing platform. In this more practical scenario of crowdsourcing, an additional challenge arises as the labels of the training instances are provided by noisy, heterogeneous crowd-workers with unknown qualities. With this motivation, we further augment our topic model to the scenario where the labels are provided by multiple noisy sources and refer to this model as ML-PA-LDA-MNS. With experiments on simulated noisy annotators, the proposed model learns the qualities of the annotators well, even with minimal training data.

The problem of automatically fixing programming errors is a very active research topic in software engineering. This is a challenging problem as fixing even a single error may require analysis of the entire program. In practice, a number of errors arise due to programmer's inexperience with the programming language or lack of attention to detail. We call these common programming errors. These are analogous to grammatical errors in natural languages. Compilers detect such errors, but their error messages are usually inaccurate. In this work, we present an end-to-end solution, called DeepFix, that can fix multiple such errors in a program without relying on any external tool to locate or fix them. At the heart of DeepFix is a multi-layered sequence-to-sequence neural network with attention which is trained to predict erroneous program locations along with the required correct statements. On a set of 6971 erroneous C programs written by students for 93 programming tasks, DeepFix could fix 1881 (27%) programs completely and 1338 (19%) programs partially.

Identifying translations from comparable corpora is a well-known problem with several applications. Existing methods rely on linguistic tools or high-quality corpora. Absence of such resources, especially in Indian languages, makes this problem hard; for example, state-of-the-art techniques achieve a mean reciprocal rank of 0.66 for English-Italian, and a mere 0.187 for Telugu-Kannada. In this work, we address the problem of comparable corpora-based translation correspondence induction (CC-TCI) when the only resources available are small noisy comparable corpora extracted from Wikipedia. We observe that translations in the source and target languages have many topically related words in common in other “auxiliary” languages. To model this, we define the notion of a translingual theme , a set of topically related words from auxiliary language corpora, and present a probabilistic framework for CC-TCI. Extensive experiments on 35 comparable corpora showed dramatic improvements in performance. We extend these ideas to propose a method for measuring cross-lingual semantic relatedness (CLSR) between words. To stimulate further research in this area, we make publicly available two new high-quality human-annotated datasets for CLSR. Experiments on the CLSR datasets show more than 200% improvement in correlation on the CLSR task. We apply the method to the real-world problem of cross-lingual Wikipedia title suggestion and build the WikiTSu system. A user study on WikiTSu shows a 20% improvement in the quality of titles suggested.

We study the problem of training an accurate linear regression model by procuring labels from multiple noisy crowd annotators, under a budget constraint. We propose a Bayesian model for linear regression in crowdsourcing and use variational inference for parameter estimation. To minimize the number of labels crowdsourced from the annotators, we adopt an active learning approach. In this specific context, we prove the equivalence of well-studied criteria of active learning like entropy minimization and expected error reduction. Interestingly, we observe that we can decouple the problems of identifying an optimal unlabeled instance and identifying an annotator to label it. We observe a useful connection between the multi-armed bandit framework and the annotator selection in active learning. Due to the nature of the distribution of the rewards on the arms, we use the Robust Upper Confidence Bound (UCB) scheme with truncated empirical mean estimator to solve the annotator selection problem. This yields provable guarantees on the regret. We further apply our model to the scenario where annotators are strategic and design suitable incentives to induce them to put in their best efforts.

Several machine learning problems arising in natural language processing can be modeled as a sequence labeling problem. Gaussian processes GPs provide a Bayesian approach to learning such problems in a kernel based framework. We develop Gaussian process models based on pseudo-likelihood to solve sequence labeling problems. The pseudo-likelihood model enables one to capture multiple dependencies among the output components of the sequence without becoming computationally intractable. We use an efficient variational Gaussian approximation method to perform inference in the proposed model. We also provide an iterative algorithm which can effectively make use of the information from the neighboring labels to perform prediction. The ability to capture multiple dependencies makes the proposed approach useful for a wide range of sequence labeling problems. Numerical experiments on some sequence labeling problems in natural language processing demonstrate the usefulness of the proposed approach.

Multi-label classification is a common supervised machine learning problem where each instance is associated with multiple classes. The key challenge in this problem is learning the correlations between the classes. An additional challenge arises when the labels of the training instances are provided by noisy, heterogeneous crowd-workers with unknown qualities. We first assume labels from a perfect source and propose a novel topic model (ML-PA-LDA) where the classes that are present as well as the classes absent generate the latent topics and hence the words. Extensive experimentation on real world datasets reveals the superior performance of the proposed model. We then non-trivially extend our topic model to the scenario where the labels are provided by noisy crowd-workers and refer to this model as ML-PA-LDA-C. With experiments on simulated crowd, the proposed model learns the qualities of the annotators well, even with minimal training data.

Due to its wide applicability, semi-supervised learning is an attractive method for using unlabeled data in classification. In this work, we present a semi-supervised support vector classifier that is designed using quasi-Newton method for nonsmooth convex functions. The proposed algorithm is suitable in dealing with very large number of examples and features. Numerical experiments on various benchmark datasets showed that the proposed algorithm is fast and gives improved generalization performance over the existing methods. Further, a non-linear semi-supervised SVM has been proposed based on a multiple label switching scheme. This non-linear semi-supervised SVM is found to converge faster and it is found to improve generalization performance on several benchmark datasets. (C) 2010 Elsevier Ltd. All rights reserved.

This letter gives an efficient algorithm for tracking the solution curve of sparse logistic regression with respect to the regularization parameter. The algorithm is based on approximating the logistic regression loss by a piecewise quadratic function, using Rosset and Zhu's path tracking algorithm on the approximate problem, and then applying a correction to get to the true path. Application of the algorithm to text classification and sparse kernel logistic regression shows that the algorithm is efficient.

Identifying translations from comparable corpora is a well-known problem with several applications, e.g. dictionary creation in resource-scarce languages. Scarcity of high quality corpora, especially in Indian languages, makes this problem hard, e.g. state-of-the-art techniques achieve a mean reciprocal rank (MRR) of 0.66 for English-Italian, and a mere 0.187 for Telugu-Kannada. There exist comparable corpora in many Indian languages with other ``auxiliary'' languages. We observe that translations have many topically related words in common in the auxiliary language. To model this, we define the notion of a translingual theme, a set of topically related words from auxiliary language corpora, and present a probabilistic framework for translation induction. Extensive experiments on 35 comparable corpora using English and French as auxiliary languages show that this approach can yield dramatic improvements in performance (e.g. MRR improves by 124% to 0.419 for Telugu-Kannada). A user study on WikiTSu, a system for cross-lingual Wikipedia title suggestion that uses our approach, shows a 20% improvement in the quality of titles suggested.

Support vector clustering involves three steps-solving an optimization problem, identification of clusters and tuning of hyper-parameters. In this paper, we introduce a pre-processing step that eliminates data points from the training data that are not crucial for clustering. Pre-processing is efficiently implemented using the R*-tree data structure. Experiments on real-world and synthetic datasets show that pre-processing drastically decreases the run-time of the clustering algorithm. Also, in many cases reduction in the number of support vectors is achieved. Further, we suggest an improvement for the step of identification of clusters.

This paper introduces a novel incremental approach to clustering interval data. The method employs rough set theory to capture the inherent uncertainty involved in cluster analysis. Our experimental results show that it produces meaningful cluster abstractions for interval data at a minimal computational expense.

In this paper a novel kernel-based soft clustering method is proposed. This method incorporates rough set theoretic flavour in support vector clustering paradigm to achieve soft clustering. Empirical studies show that this method can find soft clusters having arbitrary shapes.

Progression of hepatocellular carcinoma (HCC) is a stepwise process that proceeds from pre-neoplastic lesions—including low-grade dysplastic nodules (LGDNs) and high-grade dysplastic nodules (HGDNs)—to advanced HCC. The molecular changes associated with this progression are unclear, however, and the morphological cues thought to distinguish pre-neoplastic lesions from well-differentiated HCC are not universally accepted. To understand the multistep process of hepato-carcinogenesis at the molecular level, we used oligo-nucleotide microarrays to investigate the transcription profiles of 50 hepatocellular nodular lesions ranging from LGDNs to primary HCC (Edmondson grades 1-3). We demonstrated that gene expression profiles can discriminate not only between dysplastic nodules and overt carcinoma but also between different histological grades of HCC via unsupervised hierarchical clustering with 10,376 genes. We identified 3,084 grade-associated genes, correlated with tumor progression, using one-way ANOVA and a one-versus-all unpooled t test. Functional assignment of these genes revealed discrete expression clusters representing grade-dependent biological properties of HCC. Using both diagonal linear discriminant analysis and support vector machines, we identified 240 genes that could accurately classify tumors according to histological grade, especially when attempting to discriminate LGDNs, HGDNs, and grade 1 HCC. In conclusion, a clear molecular demarcation between dysplastic nodules and overt HCC exists. The progression from grade 1 through grade 3 HCC is associated with changes in gene expression consistent with plausible functional consequences. Supplementary material for this article can be found on the HEPATOLOGY website (http://www.interscience.wiley.com/jpages/0270-9139/suppmat/index.html). (HEPATOLOGY 2005;42:809–818.)

This paper gives a new iterative algorithm for kernel logistic regression. It is based on the solution of a dual problem using ideas similar to those of the Sequential Minimal Optimization algorithm for Support Vector Machines. Asymptotic convergence of the algorithm is proved. Computational experiments show that the algorithm is robust and fast. The algorithmic ideas can also be used to give a fast dual algorithm for solving the optimization problem arising in the inner loop of Gaussian Process classifiers.

Motivation: This paper gives a new and efficient algorithm for the sparse logistic regression problem. The proposed algorithm is based on the Gauss–Seidel method and is asymptotically convergent. It is simple and extremely easy to implement; it neither uses any sophisticated mathematical programming software nor needs any matrix operations. It can be applied to a variety of real-world problems like identifying marker genes and building a classifier in the context of cancer diagnosis using microarray data. Results: The gene selection method suggested in this paper is demonstrated on two real-world data sets and the results were found to be consistent with the literature. Availability: The implementation of this algorithm is available at the site http://guppy.mpe.nus.edu.sg/~mpessk/SparseLOGREG.shtml Supplementary Information: Supplementary material is available at the site http://guppy.mpe.nus.edu.sg/~mpessk/SparseLOGREG.shtml

Structural Support Vector Machines (SSVMs) and Conditional Random Fields (CRFs) are popular discriminative methods used for classifying structured and complex objects like parse trees, image segments and part-of-speech tags. The datasets involved are very large dimensional, and the models designed using typical training algorithms for SSVMs and CRFs are non-sparse. This non-sparse nature of models results in slow inference. Thus, there is a need to devise new algorithms for sparse SSVM and CRF classifier design. Use of elastic net and L1-regularizer has already been explored for solving primal CRF and SSVM problems, respectively, to design sparse classifiers. In this work, we focus on dual elastic net regularized SSVM and CRF. By exploiting the weakly coupled structure of these convex programming problems, we propose a new sequential alternating proximal (SAP) algorithm to solve these dual problems. This algorithm works by sequentially visiting each training set example and solving a simple subproblem restricted to a small subset of variables associated with that example. Numerical experiments on various benchmark sequence labeling datasets demonstrate that the proposed algorithm scales well. Further, the classifiers designed are sparser than those designed by solving the respective primal problems and demonstrate comparable generalization performance. Thus, the proposed SAP algorithm is a useful alternative for sparse SSVM and CRF classifier design.

Multi-task learning involves solving multiple related learning problems by sharing some common structure for improved generalization performance. A promising idea to multi-task learning is joint feature selection where a sparsity pattern is shared across task specific feature representations. In this paper, we propose a novel Gaussian Process (GP) approach to multi-task learning based on joint feature selection. The novelty of the proposed approach is that it captures the task similarity by sharing a sparsity pattern over the kernel hyper-parameters associated with each task. This is achieved by considering a hierarchical model which imposes a multi-Laplacian prior over the kernel hyper-parameters. This leads to a flexible GP model which can handle a wide range of multi-task learning problems and can identify features relevant across all the tasks. The hyper-parameter estimation results in an optimization problem which is solved using a block co-ordinate descent algorithm. Experimental results on synthetic and real world multi-task learning data sets demonstrate that the flexibility of the proposed model is useful in getting better generalization performance.

This article points out an important source of inefficiency in Platt's sequential minimal optimization (SMO) algorithm that is caused by the use of a single threshold value. Using clues from the KKT conditions for the dual problem, two threshold parameters are employed to derive modifications of SMO. These modified algorithms perform significantly faster than the original SMO on all benchmark data sets tried.

This paper points out an important source of inefficiency in Smola and Scholkopf's (1998) sequential minimal optimization (SMO) algorithm for support vector machine regression that is caused by the use of a single threshold value. Using clues from the Karush-Kuhn-Tucker conditions for the dual problem, two threshold parameters are employed to derive modifications of SMO for regression. These modified algorithms perform significantly faster than the original SMO on the datasets tried.

Ordinal regression problem arises in situations where examples are rated in an ordinal scale. In practice, labeled ordinal data are difficult to obtain while unlabeled ordinal data are available in abundance. Designing a probabilistic semi-supervised classifier to perform ordinal regression is challenging. In this work, we propose a novel approach for semi-supervised ordinal regression using Gaussian Processes (GP). It uses the expectation-propagation approximation idea, widely used for GP ordinal regression problem. The proposed approach makes use of unlabeled data in addition to the labeled data to learn a model by matching ordinal label distributions approximately between labeled and unlabeled data. The resulting mixed integer programming problem, involving model parameters (real-valued) and ordinal labels (integers) as variables, is solved efficiently using a sequence of alternating optimization steps. Experimental results on synthetic, bench-mark and real-world data sets demonstrate that the proposed GP based approach makes effective use of the unlabeled data to give better generalization performance (on the absolute error metric, in particular) than the supervised approach. Thus, it is a useful approach for probabilistic semi-supervised ordinal regression problem.

F-measure is a popular performance metric used in classification when the dataset is unbalanced. Optimizing this measure directly is often challenging since no closed form solution exists. Current algorithms use approximations to the F-measure and design classifiers using maximum margin or logistic regression framework. These algorithms are not scalable and the classifiers designed are not robust to outliers. In this work, we propose a general framework for approximate F-measure maximization. We also propose a non-convex loss function which is robust to outliers. Use of elastic net regularizer in the problem formulation enables us to do simultaneous classifier design and feature selection. We present an efficient algorithm to solve the proposed problem formulation. The proposed algorithm is simple and is easy to implement. Numerical experiments on real-world benchmark datasets demonstrate that the proposed algorithm is fast and gives better generalization performance compared to some existing approaches. Thus, it is a powerful alternative for optimizing F-measure and designing a sparse classifier.

Structural Support Vector Machines (SSVMs) have recently gained wide prominence in classifying structured and complex objects like parse-trees, image segments and Part-of-Speech (POS) tags. Typical learning algorithms used in training SSVMs result in model parameters which are vectors residing in a large-dimensional feature space. Such a high-dimensional model parameter vector contains many non-zero components which often lead to slow prediction and storage issues. Hence there is a need for sparse parameter vectors which contain a very small number of non-zero components. L1-regularizer and elastic net regularizer have been traditionally used to get sparse model parameters. Though L1-regularized structural SVMs have been studied in the past, the use of elastic net regularizer for structural SVMs has not been explored yet. In this work, we formulate the elastic net SSVM and propose a sequential alternating proximal algorithm to solve the dual formulation. We compare the proposed method with existing methods for L1-regularized Structural SVMs. Experiments on large-scale benchmark datasets show that the proposed dual elastic net SSVM trained using the sequential alternating proximal algorithm scales well and results in highly sparse model parameters while achieving a comparable generalization performance. Hence the proposed sequential alternating proximal algorithm is a competitive method to achieve sparse model parameters and a comparable generalization performance when elastic net regularized Structural SVMs are used on very large datasets.

Scatter/Gather systems are increasingly becoming useful in browsing document corpora. Usability of the present-day systems are restricted to monolingual corpora, and their methods for clustering and labeling do not easily extend to the multilingual setting, especially in the absence of dictionaries/machine translation. In this paper, we study the cluster labeling problem for multilingual corpora in the absence of machine translation, but using comparable corpora. Using a variational approach, we show that multilingual topic models can effectively handle the cluster labeling problem, which in turn allows us to design a novel Scatter/Gather system ShoBha. Experimental results on three datasets, namely the Canadian Hansards corpus, the entire overlapping Wikipedia of English, Hindi and Bengali articles, and a trilingual news corpus containing 41,000 articles, confirm the utility of the proposed system.

This paper proposes a sparse modeling approach to solve ordinal regression problems using Gaussian processes (GP). Designing a sparse GP model is important from training time and inference time viewpoints. We first propose a variant of the Gaussian process ordinal regression (GPOR) approach, leave-one-out GPOR (LOO-GPOR). It performs model selection using the leave-one-out cross-validation (LOO-CV) technique. We then provide an approach to design a sparse model for GPOR. The sparse GPOR model reduces computational time and storage requirements. Further, it provides faster inference. We compare the proposed approaches with the state-of-the-art GPOR approach on some benchmark data sets. Experimental results show that the proposed approaches are competitive.

Learning from Positive and Unlabelled examples (LPU) has emerged as an important problem in data mining and information retrieval applications. Existing techniques are not ideally suited for real world scenarios where the datasets are linearly inseparable, as they either build linear classifiers or the non-linear classifiers fail to achieve the desired performance. In this work, we propose to extend maximum margin clustering ideas and present an iterative procedure to design a non-linear classifier for LPU. In particular, we build a least squares support vector classifier, suitable for handling this problem due to symmetry of its loss function. Further, we present techniques for appropriately initializing the labels of unlabelled examples and for enforcing the ratio of positive to negative examples while obtaining these labels. Experiments on real-world datasets demonstrate that the non-linear classifier designed using the proposed approach gives significantly better generalization performance than the existing relevant approaches for LPU.

Structural Support Vector Machines (SSVMs) have become a popular tool in machine learning for predicting structured objects like parse trees, Part-of-Speech (POS) label sequences and image segments. Various efficient algorithmic techniques have been proposed for training SSVMs for large datasets. The typical SSVM formulation contains a regularizer term and a composite loss term. The loss term is usually composed of the Linear Maximum Error (LME) associated with the training examples. Other alternatives for the loss term are yet to be explored for SSVMs. We formulate a new SSVM with Linear Summed Error (LSE) loss term and propose efficient algorithms to train the new SSVM formulation using primal cutting-plane method and sequential dual coordinate descent method. Numerical experiments on benchmark datasets demonstrate that the sequential dual coordinate descent method is faster than the cutting-plane method and reaches the steady-state generalization performance faster. It is thus a useful alternative for training SSVMs when linear summed error is used.

Multi-task learning solves multiple related learning problems simultaneously by sharing some common structure for improved generalization performance of each task. We propose a novel approach to multi-task learning which captures task similarity through a shared basis vector set. The variability across tasks is captured through task specific basis vector set. We use sparse support vector machine (SVM) algorithm to select the basis vector sets for the tasks. The approach results in a sparse model where the prediction is done using very few examples. The effectiveness of our approach is demonstrated through experiments on synthetic and real multi-task datasets.

This paper proposes a novel approach to solve the ordinal regression problem using Gaussian processes. The proposed approach, probabilistic least squares ordinal regression (PLSOR), obtains the probability distribution over ordinal labels using a particular likelihood function. It performs model selection (hyperparameter optimization) using the leave-one-out cross-validation (LOO-CV) technique. PLSOR has conceptual simplicity and ease of implementation of least squares approach. Unlike the existing Gaussian process ordinal regression (GPOR) approaches, PLSOR does not use any approximation techniques for inference. We compare the proposed approach with the state-of-the-art GPOR approaches on some synthetic and benchmark data sets. Experimental results show the competitiveness of the proposed approach.

In the design of practical web page classification systems one often encounters a situation in which the labeled training set is created by choosing some examples from each class; but, the class proportions in this set are not the same as those in the test distribution to which the classifier will be actually applied. The problem is made worse when the amount of training data is also small. In this paper we explore and adapt binary SVM methods that make use of unlabeled data from the test distribution, viz., Transductive SVMs (TSVMs) and expectation regularization/constraint (ER/EC) methods to deal with this situation. We empirically show that when the labeled training data is small, TSVM designed using the class ratio tuned by minimizing the loss on the labeled set yields the best performance; its performance is good even when the deviation between the class ratios of the labeled training set and the test set is quite large. When the labeled training data is sufficiently large, an unsupervised Gaussian mixture model can be used to get a very good estimate of the class ratio in the test set; also, when this estimate is used, both TSVM and EC/ER give their best possible performance, with TSVM coming out superior. The ideas in the paper can be easily extended to multi-class SVMs and MaxEnt models.

In this paper, we develop a game theoretic approach for clustering features in a learning problem. Feature clustering can serve as an important preprocessing step in many problems such as feature selection, dimensionality reduction, etc. In this approach, we view features as rational players of a coalitional game where they form coalitions (or clusters) among themselves in order to maximize their individual payoffs. We show how Nash Stable Partition (NSP), a well known concept in the coalitional game theory, provides a natural way of clustering features. Through this approach, one can obtain some desirable properties of the clusters by choosing appropriate payoff functions. For a small number of features, the NSP based clustering can be found by solving an integer linear program (ILP). However, for large number of features, the ILP based approach does not scale well and hence we propose a hierarchical approach. Interestingly, a key result that we prove on the equivalence between a k-size NSP of a coalitional game and minimum k-cut of an appropriately constructed graph comes in handy for large scale problems. In this paper, we use feature selection problem (in a classification setting) as a running example to illustrate our approach. We conduct experiments to illustrate the efficacy of our approach.