In order to choose a neural network architecture that will be effective for aparticular modeling problem, one must understand the limitations imposed byeach of the potential options. These... | Continue reading
The interpretation of deep learning models is a challenge due to their size,complexity, and often opaque internal state. In addition, many systems, such asimage classifiers, operate on low-level... | Continue reading
We describe TF-Replicator, a framework for distributed machine learningdesigned for DeepMind researchers and implemented as an abstraction overTensorFlow. TF-Replicator simplifies writing... | Continue reading
We extend the capabilities of neural networks by coupling them to externalmemory resources, which they can interact with by attentional processes. Thecombined system is analogous to a Turing... | Continue reading
Multi-agent cooperation is an important feature of the natural world. Manytasks involve individual incentives that are misaligned with the common good,yet a wide range of organisms from bacteria... | Continue reading
From the early days of computing, games have been important testbeds forstudying how well machines can do sophisticated decision making. In recentyears, machine learning has made dramatic... | Continue reading
Due to their complex nature, it is hard to characterize the ways in whichmachine learning models can misbehave or be exploited when deployed. Recentwork on adversarial examples, i.e. inputs with... | Continue reading
We propose to meta-learn causal structures based on how fast a learner adaptsto new distributions arising from sparse distributional changes, e.g. due tointerventions, actions of agents and... | Continue reading
TensorFlow.js is a library for building and executing machine learningalgorithms in JavaScript. TensorFlow.js models run in a web browser and in theNode.js environment. The library is part of... | Continue reading
We introduce Rabbit, a combinator-based query language. Rabbit is designed tolet data analysts and other accidental programmers query complex structureddata. We combine the functional data... | Continue reading
Finding the maximum value of a function in a dynamic model plays an importantrole in many application settings, including discrete optimization in thepresence of hard constraints. We present an... | Continue reading
John Archibald Wheeler was a theoretical physicist who worked on bothdown-to-earth projects and highly speculative ideas, and always emphasized theimportance of experiment and observation, even... | Continue reading
We present a meta-learning approach for adaptive text-to-speech (TTS) withfew data. During training, we learn a multi-speaker model using a sharedconditional WaveNet core and independent learned... | Continue reading
Given a multiset $S$ of $n$ positive integers and a target integer $t$, thesubset sum problem is to decide if there is a subset of $S$ that sums up to$t$. We present a new divide-and-conquer... | Continue reading
We introduce Deep500: the first customizable benchmarking infrastructure thatenables fair comparison of the plethora of deep learning frameworks,algorithms, libraries, and techniques. The key... | Continue reading
Spreadsheets are one of the most widely used programming environments, andare widely deployed in domains like finance where errors can have catastrophicconsequences. We present a static analysis... | Continue reading
We define general linguistic intelligence as the ability to reuse previouslyacquired knowledge about a language's lexicon, syntax, semantics, and pragmaticconventions to adapt to new tasks... | Continue reading
The ability to decompose scenes in terms of abstract building blocks iscrucial for general intelligence. Where those basic building blocks sharemeaningful properties, interactions and other... | Continue reading
This paper investigates to what extent cognitive biases may affect humanunderstanding of interpretable machine learning models, in particular of rulesdiscovered from data. Twenty cognitive... | Continue reading
Recent works have highlighted the strengths of the Transformer architecturefor dealing with sequence tasks. At the same time, neural architecture searchhas advanced to the point where it can... | Continue reading
Deep neural networks are highly expressive models that have recently achievedstate of the art performance on speech and visual recognition tasks. Whiletheir expressiveness is the reason they... | Continue reading
Timing channels are a significant and growing security threat in computersystems, with no established solution. We have recently argued that the OS mustprovide time protection, in analogy to the... | Continue reading
Based on a calculation of neural decoherence rates, we argue that that thedegrees of freedom of the human brain that relate to cognitive processes shouldbe thought of as a classical rather than... | Continue reading
It has been commonly argued, on the basis of Goedel's theorem and relatedmathematical results, that true artificial intelligence cannot exist. Penrosehas further deduced from the existence of... | Continue reading
We examine the hypothesis that consciousness can be understood as a state ofmatter, "perceptronium", with distinctive information processing abilities. Weexplore five basic principles that may... | Continue reading
Model-free reinforcement learning relies heavily on a safe yet exploratorypolicy search. Proximal policy optimization (PPO) is a prominent algorithm toaddress the safe search problem, by... | Continue reading
During the total lunar eclipse of January 21, 2019 at least two meteoroidsimpacted the moon producing visible flash lights on the near side. One of theimpacts occurred on the darkest side of the... | Continue reading
Biomedical text mining has become more important than ever as the number ofbiomedical documents rapidly grows. With the progress of machine learning,extracting valuable information from... | Continue reading
Cross-validation of predictive models is the de-facto standard for modelselection and evaluation. In proper use, it provides an unbiased estimate of amodel's predictive performance. However,... | Continue reading
At the core of many important machine learning problems faced by onlinestreaming services is a need to model how users interact with the content.These problems can often be reduced to a... | Continue reading
All major web browsers now support WebAssembly, a low-level bytecode intendedto serve as a compilation target for code written in languages like C and C++.A key goal of WebAssembly is... | Continue reading
Zero-sum games such as chess and poker are, abstractly, functions thatevaluate pairs of agents, for example labeling them `winner' and `loser'. Ifthe game is approximately transitive, then... | Continue reading
Writing concurrent programs is highly error-prone due to the nondeterminismin interprocess communication. The most reliable indicators of errors inconcurrency are data races, which are accesses... | Continue reading
Wireless devices are everywhere, at home, at the office, and on the street.Devices are bombarding us with transmissions across a wide range of RFfrequencies. Many of these invisible... | Continue reading
Although deep reinforcement learning has achieved great success recently,there are still challenges in Real Time Strategy (RTS) games. Due to its largestate and action space, as well as hidden... | Continue reading
We study the connection between the highly non-convex loss function of asimple model of the fully-connected feed-forward neural network and theHamiltonian of the spherical spin-glass model under... | Continue reading
Even as end-to-end encrypted communication becomes more popular, privatemessaging remains a challenging problem due to metadata leakages, such as whois communicating with whom. Most existing... | Continue reading
This is a review of the geometry of quantum states using elementary methodsand pictures. Quantum states are represented by a convex body, often in highdimensions. In the case of n-qubits, the... | Continue reading
Creating or modifying a primary index is a time-consuming process, as theindex typically needs to be rebuilt from scratch. In this paper, we explore amore graceful "just-in-time" approach to... | Continue reading
Within the scheme of conformal cyclic cosmology (CCC), information can betransmitted from aeon to aeon. Accordingly, the "Fermi paradox" and the SETIprogramme - of communication by remote... | Continue reading
Recently, deep learning based natural language processing techniques arebeing extensively used to deal with spam mail, censorship evaluation in socialnetworks, among others. However, there is... | Continue reading
The probability distribution P from which the history of our universe issampled represents a theory of everything or TOE. We assume P is formallydescribable. Since most (uncountably many)... | Continue reading
Recently, oscillator-based Boolean computation has been proposed for itspotentials in noise immunity and energy efficiency. In such a system, logicbits are encoded in the relative phases of... | Continue reading
We present an interpretable neural network for predicting an importantclinical outcome (1-year mortality) from multi-modal Electronic Health Record(EHR) data. Our approach builds on prior... | Continue reading
Encoding information in the position of single photons has no known limits,given infinite resources. Using a heralded single-photon source and a SpatialLight Modulator (SLM), we steer single... | Continue reading
Spintronic nanodevices have ultrafast nonlinear dynamic and recurrencebehaviors on a nanosecond scale that promises to enable spintronic reservoircomputing (RC) system. Here two physical RC... | Continue reading
In recent years, the phenomenon of online misinformation and junk newscirculating on social media has come to constitute an important and widespreadproblem affecting public life online across... | Continue reading
Previous work on predicting or generating 3D human pose sequences regresseseither joint rotations or joint positions. The former strategy is prone toerror accumulation along the kinematic chain,... | Continue reading