Unit testing has been considered as having a key role in building highquality software, and therefore it has been widely used in practice. However,data on the relationship between unit testing... | Continue reading
With the capability of modeling bidirectional contexts, denoisingautoencoding based pretraining like BERT achieves better performance thanpretraining approaches based on autoregressive language... | Continue reading
The reproduction and replication of novel results has become a major issuefor a number of scientific disciplines. In computer science and relatedcomputational disciplines such as systems... | Continue reading
As the performance of computer systems stagnates due to the end of Moore'sLaw, there is a need for new models that can understand and optimize theexecution of general purpose code. While there... | Continue reading
Humans and animals show remarkable flexibility in adjusting their behaviourwhen their goals, or rewards in the environment change. While such flexibilityis a hallmark of intelligent behaviour,... | Continue reading
We propose to use a DPLL+restart to solve SAT instances by successivesimplifications based on the production of clauses that subsume the initialclauses. We show that this approach allows the... | Continue reading
We present a framework for learning to plan hierarchically in domains withunknown dynamics. We enhance planning performance by exploiting problemstructure in several ways: (i) We simplify the... | Continue reading
Dependent types help programmers write highly reliable code. However, thisreliability comes at a cost: it can be challenging to write new prototypes in(or migrate old code to) dependently-typed... | Continue reading
We present HotStuff, a leader-based Byzantine fault-tolerant replicationprotocol for the partially synchronous model. Once network communicationbecomes synchronous, HotStuff enables a correct... | Continue reading
We investigate a new commonsense inference task: given an event described ina short free-form text ("X drinks coffee in the morning"), a system reasonsabout the likely intents ("X wants to stay... | Continue reading
Visual storytelling is the task of generating stories based on a sequence ofimages. Inspired by the recent works in neural generation focusing oncontrolling the form of text, this paper explores... | Continue reading
We reflect on programming with complicated effects, recalling anundeservingly forgotten alternative to monadic programming and checking to seehow well it can actually work in modern functional... | Continue reading
Current state-of-the-art machine translation systems are based onencoder-decoder architectures, that first encode the input sequence, and thengenerate an output sequence based on the input... | Continue reading
In this article we demonstrate how to solve a variety of problems and puzzlesusing the built-in SAT solver of the computer algebra system Maple. Once theproblems have been encoded into Boolean... | Continue reading
We extend Neural Processes (NPs) to sequential data through Recurrent NPs orRNPs, a family of conditional state space models. RNPs can learn dynamicalpatterns from sequential data and deal with... | Continue reading
Learning to imitate expert behavior given action demonstrations containinghigh-dimensional, continuous observations and unknown dynamics is a difficultproblem in robotic control. Simple... | Continue reading
While adversarial training can improve robust accuracy (against anadversary), it sometimes hurts standard accuracy (when there is no adversary).Previous work has studied this tradeoff between... | Continue reading
Nanomechanical computers promise a greatly improved energetic efficiencycompared to their electrical counterparts. However, progress towards this goalis hindered by a lack of modular components,... | Continue reading
Smaller, smarter and faster edge devices in the Internet of things erademands secure data analysis and transmission under resource constraints ofhardware architecture. Lightweight cryptography... | Continue reading
Recent studies have shown that many results published in peer-reviewedscientific journals are not reproducible. This raises the following question:why is it so easy to fool myself into believing... | Continue reading
This paper tackles the problem of open domain factual Arabic questionanswering (QA) using Wikipedia as our knowledge source. This constrains theanswer of any question to be a span of text in... | Continue reading
People can learn a new concept and use it compositionally, understanding howto "blicket twice" after learning how to "blicket." In contrast, powerfulsequence-to-sequence (seq2seq) neural... | Continue reading
We present the first comprehensive study on automatic knowledge baseconstruction for two prevalent commonsense knowledge graphs: ATOMIC (Sap etal., 2019) and ConceptNet (Speer et al., 2017).... | Continue reading
This note argues for more use of simple models beyond Amdahl's Law:Bottleneck Analysis, Little's Law, and a M/M/1 Queue. | Continue reading
We present FAKTA which is a unified framework that integrates variouscomponents of a fact checking process: document retrieval from media sourceswith various types of reliability, stance... | Continue reading
Since 1950, when Alan Turing proposed what has since come to be called theTuring test, the ability of a machine to pass this test has established itselfas the primary hallmark of general AI. To... | Continue reading
Fine-grained Entity Typing is a tough task which suffers from noise samplesextracted from distant supervision. Thousands of manually annotated samples canachieve greater performance than... | Continue reading
We incorporate morphological supervision into character language models(CLMs) via multitasking and show that this addition improves bits-per-character(BPC) performance across 24 languages, even... | Continue reading
Human decision-making underlies all economic behavior. For the past fourdecades, human decision-making under uncertainty has continued to be explainedby theoretical models based on prospect... | Continue reading
This book introduces a temporal type theory, the first of its kind as far aswe know. It is based on a standard core, and as such it can be formalized in aproof assistant such as Coq or Lean by... | Continue reading
We employ the observed Gaia proper motions of the newly discovered Antlia 2dwarf galaxy to calculate its orbital distribution in the cosmologically recentpast. Using these observationally... | Continue reading
The Standard Model violates parity, but only by mechanisms which areinvisible to Large Hadron Collider (LHC) experiments (on account of the lack ofinitial state polarisation or spin-sensitivity... | Continue reading
We consider the task of mapping pseudocode to long programs that arefunctionally correct. Given test cases as a mechanism to validate programs, wesearch over the space of possible translations... | Continue reading
Sarcasm is often expressed through several verbal and non-verbal cues, e.g.,a change of tone, overemphasis in a word, a drawn-out syllable, or a straightlooking face. Most of the recent work in... | Continue reading
Climate change is one of the greatest challenges facing humanity, and we, asmachine learning experts, may wonder how we can help. Here we describe howmachine learning can be a powerful tool in... | Continue reading
This note is a critical examination of the argument of Frauchiger and Renner(Nature Communications 9:3711 (2018)), in which they claim to show that threereasonable assumptions about the use of... | Continue reading
Recent works have shown the effectiveness of randomized smoothing as ascalable technique for building neural network-based classifiers that areprovably robust to $\ell_2$-norm adversarial... | Continue reading
The human ability to imagine the variety of appearances of novel objectsbased on past experience is crucial for quickly learning novel visual conceptsbased on few examples. Endowing machines... | Continue reading
Generative Adversarial Networks (GANs) learn to model data distributionsthrough two unsupervised neural networks, each minimizing the objectivefunction maximized by the other. We relate this... | Continue reading
We explore the use of Vector Quantized Variational AutoEncoder (VQ-VAE)models for large scale image generation. To this end, we scale and enhance theautoregressive priors used in VQ-VAE to... | Continue reading
Large pre-trained neural networks such as BERT have had great recent successin NLP, motivating a growing body of research investigating what aspects oflanguage they are able to learn from... | Continue reading
Model-based Reinforcement Learning approaches have the promise of beingsample efficient. Much of the progress in learning dynamics models in RL hasbeen made by learning models via supervised... | Continue reading
While the evidence for dark matter continues to grow, the nature of the darkmatter remains a mystery. A dark $U(1)_D$ gauge theory can have a small kineticmixing with the visible photon which... | Continue reading
We have developed a theory for the contact between suction cups and randomlyrough surfaces. The theory predicts the dependency of the pull-off time(lifetime) on the pull-off force, and is tested... | Continue reading
Imitation learning allows agents to learn complex behaviors fromdemonstrations. However, learning a complex vision-based task may require animpractical number of demonstrations. Meta-imitation... | Continue reading
Human speech is often accompanied by hand and arm gestures. Given audiospeech input, we generate plausible gestures to go along with the sound.Specifically, we perform cross-modal translation... | Continue reading
Disassembly is fundamental to binary analysis and rewriting. We present anovel disassembly technique that takes a stripped binary and producesreassembleable assembly code. The resulting assembly... | Continue reading
Recent progress in hardware and methodology for training neural networks hasushered in a new generation of large networks trained on abundant data. Thesemodels have obtained notable gains in... | Continue reading