Speech applications dealing with conversations require not only recognizingthe spoken words, but also determining who spoke when. The task of assigningwords to speakers is typically addressed by... | Continue reading
Relation extraction (RE) aims at extracting the relation between two entitiesfrom the text corpora. It is a crucial task for Knowledge Graph (KG)construction. Most existing methods predict the... | Continue reading
Exploratory data analysis is a fundamental aspect of knowledge discovery thataims to find the main characteristics of a dataset. Dimensionality reduction,such as manifold learning, is often used... | Continue reading
We introduce our efforts towards building a universal neural machinetranslation (NMT) system capable of translating between any language pair. Weset a milestone towards this goal by building a... | Continue reading
Not all types of supervision signals are created equal: Different types offeedback have different costs and effects on learning. We show howself-regulation strategies that decide when to ask for... | Continue reading
Building computer systems that can converse about their visual environment isone of the oldest concerns of research in Artificial Intelligence andComputational Linguistics (see, for example,... | Continue reading
Neural network-based generative language models like ELMo and BERT can workeffectively as general purpose sentence encoders in text classification withoutfurther fine-tuning. Is it possible to... | Continue reading
Many real-world human behaviors can be characterized as a sequential decisionmaking processes, such as urban travelers choices of transport modes and routes(Wu et al. 2017). Differing from... | Continue reading
Model-free reinforcement learning (RL) can be used to learn effectivepolicies for complex tasks, such as Atari games, even from image observations.However, this typically requires very large... | Continue reading
Many multiagent applications require an agent to learn quickly how tointeract with previously unknown other agents. To address this problem,researchers have studied learning algorithms which... | Continue reading
Robots navigating in human environments should use language to ask forassistance and be able to understand human responses. To study this challenge,we introduce Cooperative Vision-and-Dialog... | Continue reading
Topic modeling analyzes documents to learn meaningful patterns of words.However, existing topic models fail to learn interpretable topics when workingwith large and heavy-tailed vocabularies. To... | Continue reading
Deep neural networks (DNNs) are found to be vulnerable against adversarialexamples, which are carefully crafted inputs with a small magnitude ofperturbation aiming to induce arbitrarily... | Continue reading
Reflecting on the advances of off-policy deep reinforcement learning (RL)algorithms since the development of DQN in 2013, it is important to ask: arethe complexities of recent off-policy methods... | Continue reading
We have developed the concept of pathway assembly to explore the amount ofextrinsic information required to build an object. To quantify this informationin an agnostic way, we present a method... | Continue reading
Probabilistic logic reasoning is a central component of such cognitivearchitectures as OpenCog. However, as an integrative architecture, OpenCogfacilitates cognitive synergy via hybridization of... | Continue reading
We demonstrate the possibility of what we call sparse learning: acceleratedtraining of deep neural networks that maintain sparse weights throughouttraining while achieving performance levels... | Continue reading
In this paper we will prove that GCM-models used in IPCC report AR5 fail tocalculate the influences of the low cloud cover changes on the globaltemperature. That is why those models give a very... | Continue reading
Recommendation problems with large numbers of discrete items, such asproducts, webpages, or videos, are ubiquitous in the technology industry. Deepneural networks are being increasingly used for... | Continue reading
Neural parsers obtain state-of-the-art results on benchmark treebanks forconstituency parsing -- but to what degree do they generalize to other domains?We present three results about the... | Continue reading
There is extensive, yet fragmented, evidence of gender differences inacademia suggesting that women are under-represented in most scientificdisciplines, publish fewer articles throughout a... | Continue reading
It can be challenging to train multi-task neural networks that outperform oreven match their single-task counterparts. To help address this, we proposeusing knowledge distillation where... | Continue reading
We present a scheme for sequential decision making with a risk-sensitiveobjective and constraints in a dynamic environment. A neural network is trainedas an approximator of the mapping from... | Continue reading
Artificial Intelligence models are becoming increasingly more powerful andaccurate, supporting or even replacing humans' decision making. But withincreased power and accuracy also comes higher... | Continue reading
In complex scenarios where a model of other actors is necessary to predictand interpret their actions, it is often desirable that the model works wellwith a wide variety of previously unknown... | Continue reading
We present a new architecture for storing and accessing entity mentionsduring online text processing. While reading the text, entity references areidentified, and may be stored by either... | Continue reading
Transformer architectures show significant promise for natural languageprocessing. Given that a single pretrained model can be fine-tuned to performwell on many different tasks, these networks... | Continue reading
We consider general aspects of the physics underlying the flight of Gossamerspiders, also known as ballooning spiders. We show that existing observationsand the physics of spider silk in the... | Continue reading
In this paper, we prove that depth with nonlinearity creates no bad localminima in a type of arbitrarily deep ResNets with arbitrary nonlinearactivation functions, in the sense that the values... | Continue reading
We introduce NoisyNet, a deep reinforcement learning agent with parametricnoise added to its weights, and show that the induced stochasticity of theagent's policy can be used to aid efficient... | Continue reading
Metalens as one of the most popular applications of emmerging opticalmetasurfaces has raised widspread interest recently. With nano structures fullycontrolling phase, polarization and... | Continue reading
Background: Continuous Integration (CI) systems are now the bedrock ofseveral software development practices. Several tools such as TravisCI,CircleCI, and Hudson, that implement CI practices,... | Continue reading
We propose the Fréchet Audio Distance (FAD), a novel, reference-freeevaluation metric for music enhancement algorithms. We demonstrate how typicalevaluation metrics for speech enhancement and... | Continue reading
We introduce the Neural State Machine, seeking to bridge the gap between theneural and symbolic views of AI and integrate their complementary strengths forthe task of visual reasoning. Given an... | Continue reading
Machine Translation (MT) is an area in natural language processing, whichfocus on translating from one language to another. Many approaches ranging fromstatistical methods to deep learning... | Continue reading
Our entry into the HAHA 2019 Challenge placed $3^{rd}$ in the classificationtask and $2^{nd}$ in the regression task. We describe our system andinnovations, as well as comparing our results to a... | Continue reading
Cross-lingual embeddings aim to represent words in multiple languages in ashared vector space by capturing semantic similarities across languages. Theyare a crucial component for scaling tasks... | Continue reading
In this paper, we make freely accessible ANETAC our English-Arabic namedentity transliteration and classification dataset that we built from freelyavailable parallel translation corpora. The... | Continue reading
This paper introduces \texttt{infotheory}: a package written in C++ andusable from Python and C++, for multivariate information theoretic analyses ofdiscrete and continuous data. This package... | Continue reading
The scientific study of consciousness is a new field which has emerged as aresponse to groundbreaking developments in neuroscience, cognitive psychologyand analytic philosophy. Its aim is to... | Continue reading
Machine learning approaches for building task-oriented dialogue systemsrequire large conversational datasets with labels to train on. We areinterested in building task-oriented dialogue systems... | Continue reading
We present highly scalable parallel distributed-memory algorithms andassociated data structures for a generic finite element framework that supports$h$-adaptivity on computational domains... | Continue reading
The automotive industry has seen an increased need for connectivity, both asa result of the advent of autonomous driving and the rise of connected cars andtruck fleets. This shift has led to... | Continue reading
When implementing secure software, developers must ensure certainrequirements, such as the erasure of secret data after its use and execution inreal time. Such requirements are not explicitly... | Continue reading
Intent Detection and Slot Filling are two pillar tasks in Spoken NaturalLanguage Understanding. Common approaches adopt joint Deep Learningarchitectures in attention-based recurrent frameworks.... | Continue reading
Head-driven phrase structure grammar (HPSG) enjoys a uniform formalismrepresenting rich contextual syntactic and even semantic meanings. This papermakes the first attempt to formulate a... | Continue reading
Mental illness affects a significant portion of the worldwide population.Online mental health forums can provide a supportive environment for thoseafflicted and also generate a large amount of... | Continue reading
Procedural generation of initial states of state-space search problems haveapplications in human and machine learning as well as in the evaluation ofplanning systems. In this paper we deal with... | Continue reading