Several real-world applications require real-time prediction on resource-scarce devices such as an Internet of Things (IoT) sensor. Such applications demand prediction models with small storage and computational complexity that do not compromise significantly on accuracy. In this work, we propose ProtoNN, a novel algorithm that addresses the problem of real-time and accurate prediction on resource-scarce devices. ProtoNN is inspired by k-Nearest Neighbor (KNN) but has several orders lower storage and prediction complexity. ProtoNN models can be deployed even on devices with puny storage and computational power (e.g. an Arduino UNO with 2kB RAM) to get excellent prediction accuracy. ProtoNN derives its strength from three key ideas: a) learning a small number of prototypes to represent the entire training set, b) sparse low dimensional projection of data, c) joint discriminative learning of the projection and prototypes with explicit model size constraint. We conduct systematic empirical evaluation of ProtoNN on a variety of supervised learning tasks (binary, multi-class, multi-label classification) and show that it gives nearly state-of-the-art prediction accuracy on resource-scarce devices while consuming several orders lower storage, and using minimal working memory.

Collections that contain a large number of short texts are becoming increasingly common (eg., tweets, reviews, etc). Analytical tasks (such as classification, clustering, etc.) involving short texts could be challenging due to the lack of context and owing to their sparseness. An often encountered problem is low accuracy on the task. A standard technique used in the handling of short texts is expanding them before subjecting them to the task. However, existing works on short text expansion suffer from certain limitations: (i) they depend on domain knowledge to expand the text; (ii) they employ task-specific heuristics; and (iii) the expansion procedure is tightly coupled to the task. This makes it hard to adapt a procedure, designed for one task, into another. We present an expansion technique -- TIDE (Task-specIfic short Document Expansion) -- that can be applied on several Machine Learning, NLP and Information Retrieval tasks on short texts (such as short text classification, clustering, entity disambiguation, and the like) without using task specific heuristics and domain-specific knowledge for expansion. At the same time, our technique is capable of learning to expand short texts in a task-specific way. That is, the same technique that is applied to expand a short text in two different tasks is able to learn to produce different expansions depending upon what expansion benefits the task's performance. To speed up the learning process, we also introduce a technique called block learning. Our experiments with classification and clustering tasks show that our framework improves upon several baselines according to the standard evaluation metrics which includes the accuracy and normalized mutual information (NMI).

Emails continue to remain the most important and widely used mode of online communication despite having its origins in the middle of last century and being threatened by a variety of online communication innovations. While several studies have predicted the continuous growth of volume of email communication, there is little innovation on improving the search in emails, an imperative part of the user experience. In this work, we present a lightweight email application codenamed InLook, that intends to provide a productive search experience.

We consider the problem of providing spelling corrections for misspelled queries in Email Search using user’s own mail data. A popular strategy for general query spelling correction is to generate corrections from query logs. However, this strategy is not effective in Email Search for two reasons: 1) query log of any sin- gle user is typically not rich enough to provide potential corrections for a new query 2) corrections generated us- ing query logs of other users are not particularly useful since the mail data as well as search intent are highly specific to the user. We address the challenge of design- ing an effective spelling correction algorithm for Email Search in the absence of query logs. We propose SpEQ, a Machine Learning based approach that generates cor- rections for misspelled queries directly from the user’s own mail data.

We address the problem of mining name transliterations from comparable corpora in languages P and Q in the following resource-poor scenario:Parallel names in PQ are not available for training. Parallel names in PR and RQ are available for training.We propose a novel solution for the problem by computing a common geometric feature space for P,Q and R where name transliterations are mapped to similar vectors. We employ Canonical Correlation Analysis (CCA) to compute the common geometric feature space using only parallel names in PR and RQ and without requiring parallel names in PQ. We test our algorithm on data sets in several languages and show that it gives results comparable to the state-of-the-art transliteration mining algorithms that use parallel names in PQ for training.

People Search is an important search service with multiple applications (eg. looking up a friend on Facebook, finding colleagues in corporate email directories etc). With the proportion of non-English users on a steady rise, people search services are being used by users from diverse language demographics. Users may issue name search queries against these directories in languages other than the language of the directory, in which case the present monolingual name search approaches will not work. In this demo, we present a Multilingual People Search system capable of performing fast name lookups on large user directories, independent of the directory language. Our system has applications in areas like social networking, enterprise search and email address book search.

Suggesting topics that are related to user's goal or interest is very important in web search. However, search engines today focus on suggesting mainly reformulations and lexical variants of the query mined from query logs. In this demonstration, we show a system that can suggest related topics for a query based on the top search results for the query. It can help users in exploring the topics related to their information need. The topic suggestion system can be integrated with any search engine or it can be easily installed on the client machine as a browser plugin.

Although Pseudo-Relevance Feedback (PRF) techniques improve average retrieval performance at the price of high variance, not much is known about their optimality and the reasons for their instability. In this work, we study more than 800 topics from several test collections including the TREC Robust Track and show that PRF techniques are highly suboptimal, i.e. they do not make the fullest utilization of pseudo-relevant documents and under-perform. A careful selection of expansion terms from the pseudo-relevant document with the help of an oracle can actually improve retrieval performance dramatically (by > 60%). Further, we show that instability in PRF techniques is mainly due to wrong selection of expansion terms from the pseudo-relevant documents. Our findings emphasize the need to revisit the problem of term selection to make a break through in PRF.

No abstract available.

It is desirable to enhance the fingerprint image for achieving good fingerprint matching performance. In our earlier work, we have proposed a method for learning a set of partitioned least-squares filters from a given set of images and ground truth pairs. In this paper, we propose a refined approach to learn a class of hierarchical filter banks to extend our enhancement technique. The new technique is shown to obtain much better results particularly when the input image is not easily restored using our earlier technique. We evaluate the performance of our new approach to fingerprint enhancement filter design by assessing the effect of enhancement on performance of (i) fingerprint feature extraction and (ii) fingerprint matching.

Fingerprint images contain varying amount of noise because of the limitations of the fingerprint acquisition process. It is often necessary to enhance such noisy fingerprint images so that the features extracted from them are reliable. We propose a novel approach to fingerprint enhancement where a set of filters are learned using the "learn-from-example" paradigm. An expert provides the ground truth information for ridges in a small set of representative fingerprint images. The space of local fingerprint patterns in a small neighborhood is partitioned into a set of expressive yet computationally simple classes. A filter is learnt for each partition by finding the optimal linear mapping (in least-square sense) from the input to the enhanced space. The proposed approach offers distinct performance and speed advantages for a wide variety of fingerprint images.