Drivers are written in C or restricted subsets of C++ on all production-gradeserver, desktop, and mobile operating systems. They account for 66% of the codein Linux, but 39 out of 40 security... | Continue reading
It has been claimed and disputed that World War II has been followed by a`long peace', an unprecedented decline of war. We conduct a full changepointanalysis of well-documented,... | Continue reading
We outline the idiosyncrasies of neural information processing and machinelearning in quantitative finance. We also present some of the approaches wetake towards solving the fundamental... | Continue reading
The field of edge and fog computing is growing, but there are still manyinconsistent and loosely-defined terms in current literature. With manyarticles comparing theoretical architectures and... | Continue reading
Most existing machine learning classifiers are highly vulnerable toadversarial examples. An adversarial example is a sample of input data whichhas been modified very slightly in a way that is... | Continue reading
The air transportation network, a fundamental component of criticalinfrastructure, is formed from a collection of individual air carriers, eachone with a methodically designed and engineered... | Continue reading
DevOps is a collaborative and multidisciplinary organizational effort toautomate continuous delivery of new software updates while guaranteeing theircorrectness and reliability. The present... | Continue reading
A hallmark of human intelligence is the ability to understand and communicatewith language. Interactive Fiction games are fully text-based simulationenvironments where a player issues text... | Continue reading
Previous storytelling approaches mostly focused on optimizing traditionalmetrics such as BLEU, ROUGE and CIDEr. In this paper, we re-examine thisproblem from a different angle, by looking deep... | Continue reading
Vehicle 3D extents and trajectories are critical cues for predicting thefuture location of vehicles and planning future agent ego-motion based on thosepredictions. In this paper, we propose a... | Continue reading
Pretrained language models are promising particularly for low-resourcelanguages as they only require unlabelled data. However, training existingmodels requires huge amounts of compute, while... | Continue reading
The biomedical scientific literature comprises a crucial, sometimeslife-saving, natural language resource whose size is accelerating over time.The information in this resource tends to follow a... | Continue reading
Word embedding models such as the skip-gram learn vector representations ofwords' semantic relationships, and document embedding models learn similarrepresentations for documents. On the other... | Continue reading
We propose an approach to estimate the 6DOF pose of a satellite, relative toa canonical pose, from a single image. Such a problem is crucial in many spaceproximity operations, such as docking,... | Continue reading
We analyze gravitational-wave data from the first LIGO detection of a binaryblack-hole merger (GW150914) in search of the ringdown of the remnant blackhole. Using observations beginning at the... | Continue reading
"PL/SQL functions are slow," is common developer wisdom that derives from thetension between set-oriented SQL evaluation and statement-by-statement PL/SQLinterpretation. We pursue the radical... | Continue reading
Prior work has proposed effective methods to learn event representations thatcan capture syntactic and semantic information over text corpus, demonstratingtheir effectiveness for downstream... | Continue reading
Humans are capable of abstracting away irrelevant details when studyingproblems. This is especially noticeable for problems over grid-cells, as humansare able to disregard certain parts of the... | Continue reading
Many question answering (QA) tasks only provide weak supervision for how theanswer should be computed. For example, TriviaQA answers are entities that canbe mentioned multiple times in... | Continue reading
Transferring high-level knowledge from a source task to a target task is aneffective way to expedite reinforcement learning (RL). For example,propositional logic and first-order logic have been... | Continue reading
Current training regimes for deep learning usually involve exposure to asingle task / dataset at a time. Here we start from the observation that inthis context the trained model is not given any... | Continue reading
BERT (Bidirectional Encoder Representations from Transformers) and relatedpre-trained Transformers have provided large gains across many languageunderstanding tasks, achieving a new... | Continue reading
The use of semantic descriptions in data intensive domains require asystematic model for linking semantic descriptions with their manifestations infragments of heterogeneous information and data... | Continue reading
Equal access to voting is a core feature of democratic government. Using datafrom millions of smartphone users, we quantify a racial disparity in votingwait times across a nationwide sample of... | Continue reading
We propose self-teaching networks to improve the generalization capacity ofdeep neural networks. The idea is to generate soft supervision labels using theoutput layer for training the lower... | Continue reading
Neural machine translation - using neural networks to translate humanlanguage - is an area of active research exploring new neuron types and networktopologies with the goal of dramatically... | Continue reading
Neural networks are part of many contemporary NLP systems, yet theirempirical successes come at the price of vulnerability to adversarial attacks.Previous work has used adversarial training and... | Continue reading
MIT researchers have found new ways to cure headaches in manufacturing carbon nanotube processors, which are faster and less power hungry than silicon chips. | Continue reading
A core capability of intelligent systems is the ability to quickly learn newtasks by drawing on prior experience. Gradient (or optimization) basedmeta-learning has recently emerged as an... | Continue reading
Contextual word representations, typically trained on unstructured, unlabeledtext, do not contain any explicit grounding to real world entities and areoften unable to remember facts about those... | Continue reading
We propose a novel architecture which is able to automatically anonymizefaces in images while retaining the original data distribution. We ensure totalanonymization of all faces in an image by... | Continue reading
Bipartite b-matching, where agents on one side of a market are matched to oneor more agents or items on the other, is a classical model that is used inmyriad application areas such as... | Continue reading
Knowledge graphs are important resources for many artificial intelligencetasks but often suffer from incompleteness. In this work, we propose to usepre-trained language models for knowledge... | Continue reading
We present a system that automatically discovers critical mechanics in avariety of video games within the General Video Game Artificial Intelligence(GVG-AI) framework using a combination of game... | Continue reading
Probes, supervised models trained to predict properties (likeparts-of-speech) from representations (like ELMo), have achieved high accuracyon a range of linguistic tasks. But does this mean that... | Continue reading
Define the \emph{Collatz map} $\mathrm{Col} : \mathbb{N}+1 \to \mathbb{N}+1$on the positive integers $\mathbb{N}+1 = \{1,2,3,\dots\}$ by setting$\mathrm{Col}(N)$ equal to $3N+1$ when $N$ is odd... | Continue reading
Perhaps the most ambitious scientific quest in human history is the creationof general artificial intelligence, which roughly means AI that is as smart orsmarter than humans. The dominant... | Continue reading
We present a method to produce abstractive summaries of long documents thatexceed several thousand words via neural abstractive summarization. We performa simple extractive step before... | Continue reading
Recently, chatbots received an increased attention from industry and diverseresearch communities as a dialogue-based interface providing advancedhuman-computer interactions. On the other hand,... | Continue reading
Lecture notes on optimization for machine learning, derived from a course atPrinceton University and tutorials given in MLSS, Buenos Aires, as well asSimons Foundation, Berkeley. | Continue reading
Catastrophic forgetting and capacity saturation are the central challenges ofany parametric lifelong learning system. In this work, we study thesechallenges in the context of sequential... | Continue reading
Text compression has diverse applications such as Summarization, ReadingComprehension and Text Editing. However, almost all existing approaches requireeither hand-crafted features, syntactic... | Continue reading
Measurements of 21 cm Epoch of Reionization structure are subject tosystematics originating from both the analysis and the observation conditions.Using 2013 data from the Murchison Widefield... | Continue reading
TorchGAN is a PyTorch based framework for writing succinct and comprehensiblecode for training and evaluation of Generative Adversarial Networks. Theframework's modular design allows effortless... | Continue reading
In order to facilitate the accesses of general users to knowledge graphs, anincreasing effort is being exerted to construct graph-structured queries ofgiven natural language questions. At the... | Continue reading
This paper formally compares some central notions from two well-knownformalisms for rule-based argumentation, DeLP and ASPIC+. The comparisonsespecially focus on intuitive adequacy and... | Continue reading
A lot of efforts have been devoted to investigating how agents can learneffectively and achieve coordination in multiagent systems. However, it isstill challenging in large-scale multiagent... | Continue reading