A number of recent efforts have attempted to design accelerators for popular machine learning algorithms, such as those involving convolutional and deep neural networks (CNNs DNNs). These algorithms typically involve a large multiply-accumulate (dot-product) operations. project, DaDianNao, adopts near data processing approach, where specialized functional unit performs all the digital arithmetic operations receives input weights from adjacent eDRAM banks. This work explores an in-situ...

Anomaly detection is a critical step towards building secure and trustworthy system. The primary purpose of system log to record states significant events at various points help debug failures perform root cause analysis. Such data universally available in nearly all computer systems. Log an important valuable resource for understanding status performance issues; therefore, the logs are naturally excellent source information online monitoring anomaly detection. We propose DeepLog, deep...

A number of recent efforts have attempted to design accelerators for popular machine learning algorithms, such as those involving convolutional and deep neural networks (CNNs DNNs). These algorithms typically involve a large multiply-accumulate (dot-product) operations. project, DaDianNao, adopts near data processing approach, where specialized functional unit performs all the digital arithmetic operations receives input weights from adjacent eDRAM banks. This work explores an in-situ...

Jonathan Berant, Vivek Srikumar, Pei-Chun Chen, Abby Vander Linden, Brittany Harding, Brad Huang, Peter Clark, Christopher D. Manning. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP). 2014.

Constructing distributed representations for words through neural language models and using the resulting vector spaces analysis has become a crucial component of natural processing (NLP). However, despite their widespread application, little is known about structure properties these spaces. To gain insights into relationship between words, NLP community begun to adapt high-dimensional visualization techniques. In particular, researchers commonly use t-distributed stochastic neighbor...

Word embeddings carry stereotypical connotations from the text they are trained on, which can lead to invalid inferences in downstream models that rely on them. We use this observation design a mechanism for measuring stereotypes using task of natural language inference. demonstrate reduction via bias mitigation strategies static word (GloVe). Further, we show gender bias, these techniques extend contextualized when applied selectively only components (ELMo, BERT).

In this paper, we observe that semi-structured tabulated text is ubiquitous; understanding them requires not only comprehending the meaning of fragments, but also implicit relationships between them. We argue such data can prove as a testing ground for how reason about information. To study this, introduce new dataset called INFOTABS, comprising human-written textual hypotheses based on premises are tables extracted from Wikipedia info-boxes. Our analysis shows semi-structured, multi-domain...

Today, the dominant paradigm for training neural networks involves minimizing task loss on a large dataset. Using world knowledge to inform model, and yet retain ability perform end-to-end remains an open question. In this paper, we present novel framework introducing declarative network architectures in order guide prediction. Our systematically compiles logical statements into computation graphs that augment without extra learnable parameters or manual redesign. We evaluate our modeling...

Tao Li, Vivek Gupta, Maitrey Mehta, Srikumar. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

Nathan Schneider, Jena D. Hwang, Vivek Srikumar, Jakob Prange, Austin Blodgett, Sarah R. Moeller, Aviram Stern, Adi Bitan, Omri Abend. Proceedings of the 56th Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2018.

Background: Training therapists is both expensive and time-consuming. Degree–based training can require tens of thousands dollars hundreds hours expert instruction. Counseling skills practice often involves role-plays, standardized patients, or with real clients. Performance–based feedback critical for skill development expertise, but trainee receive minimal subjective feedback, which distal to their practice. Objective: In this study, we developed evaluated a patient-like neural...

This paper introduces the problem of predicting semantic relations expressed by prepositions and develops statistical learning models for relations, their arguments types arguments. We define an inventory 32 building on word sense disambiguation task collapsing related senses across prepositions. Given a preposition in sentence, our computational to jointly model relation its along with types, as way support prediction. The annotated data, however, only provides labels label, not types....

Many recent works take advantage of highly parallel analog in-situ computation in memristor crossbars to accelerate the many vector-matrix multiplication operations deep neural networks (DNNs). However, these accelerators have two significant shortcomings: The ADCs account for a large fraction chip power and area, adopt homogeneous design which every resource is provisioned worst case. By addressing both problems, new architecture, called Newton, moves closer achieving optimal energy per...

With the recent advances in deep learning, neural network models have obtained state-of-the-art performances for many linguistic tasks natural language processing. However, this rapid progress also brings enormous challenges. The opaque nature of a model leads to hard-to-debug-systems and difficult-to-interpret mechanisms. Here, we introduce visualization system that, through tight yet flexible integration between elements underlying model, allows user interrogate by perturbing input,...

While language embeddings have been shown to stereotyping biases, how these biases affect downstream question answering (QA) models remains unexplored. We present UNQOVER, a general framework probe and quantify through underspecified questions. show that naive use of model scores can lead incorrect bias estimates due two forms reasoning errors: positional dependence independence. design formalism isolates the aforementioned errors. As case studies, we this metric analyze four important...

Ashim Gupta, Vivek Srikumar. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). 2021.

Reasoning about tabular information presents unique challenges to modern NLP approaches which largely rely on pre-trained contextualized embeddings of text. In this paper, we study these through the problem natural language inference. We propose easy and effective modifications how is presented a model for task. show via systematic experiments that strategies substantially improve inference performance.

Deep learning has shown recent success in classifying anomalies chest x-rays, but datasets are still small compared to natural image datasets. Supervision of abnormality localization been improve trained models, partially compensating for dataset sizes. However, explicitly labeling these requires an expert and is very time-consuming. We propose a potentially scalable method collecting implicit data using eye tracker capture gaze locations microphone dictation report, imitating the setup...

English prepositions are extremely frequent and extraordinarily polysemous. In some usages they contribute information about spatial, temporal, or causal roles/relations; in other cases institutionalized, somewhat arbitrarily, as case markers licensed by a particular governing verb, verb class, syntactic construction. To facilitate automatic disambiguation, we propose general-purpose, broadcoverage taxonomy of preposition functions that call supersenses: these coarse unlexicalized so to be...