Results for "Demis Hassabis"

total 28took 0.04s
Approximate Hubel-Wiesel Modules and the Data Structures of Neural ComputationDec 28 2015This paper describes a framework for modeling the interface between perception and memory on the algorithmic level of analysis. It is consistent with phenomena associated with many different brain regions. These include view-dependence (and invariance) ... More
Noisy Networks for ExplorationJun 30 2017Feb 15 2018We introduce NoisyNet, a deep reinforcement learning agent with parametric noise added to its weights, and show that the induced stochasticity of the agent's policy can be used to aid efficient exploration. The parameters of the noise are learned with ... More
Memory-based Parameter AdaptationFeb 28 2018Deep neural networks have excelled on a wide range of problems, from vision to language and game playing. Neural networks very gradually incorporate information into weights as they process data, requiring very low learning rates. If the training distribution ... More
Learning and Querying Fast Generative Models for Reinforcement LearningFeb 08 2018A key challenge in model-based reinforcement learning (RL) is to synthesize computationally efficient and accurate environment models. We show that carefully designed generative models that learn and operate on compact state representations, so-called ... More
SCAN: Learning Hierarchical Compositional Visual ConceptsJul 11 2017Jun 06 2018The seemingly infinite diversity of the natural world arises from a relatively small set of coherent rules, such as the laws of physics or chemistry. We conjecture that these rules give rise to regularities that can be discovered through primarily unsupervised ... More
Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning AlgorithmDec 05 2017The game of chess is the most widely-studied domain in the history of artificial intelligence. The strongest programs are based on a combination of sophisticated search techniques, domain-specific adaptations, and handcrafted evaluation functions that ... More
Grounded Language Learning in a Simulated 3D WorldJun 20 2017Jun 26 2017We are increasingly surrounded by artificially intelligent technology that takes decisions and executes actions on our behalf. This creates a pressing need for general means to communicate with, instruct and guide artificial agents, with human language ... More
Overcoming catastrophic forgetting in neural networksDec 02 2016The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Neural networks are not, in general, capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist ... More
Imagination-Augmented Agents for Deep Reinforcement LearningJul 19 2017Feb 14 2018We introduce Imagination-Augmented Agents (I2As), a novel architecture for deep reinforcement learning combining model-free and model-based aspects. In contrast to most existing model-based reinforcement learning and planning methods, which prescribe ... More
A General Mass Transference PrincipleMar 07 2018Aug 17 2018In this paper we prove a general form of the Mass Transference Principle for $\limsup$ sets defined via neighbourhoods of sets satisfying a certain local scaling property. Such sets include self-similar sets satisfying the open set condition and smooth ... More
The Mass Transference Principle: Ten Years OnApr 21 2017May 09 2017In this article we discuss the Mass Transference Principle due to Beresnevich and Velani and survey several generalisations and variants, both deterministic and random. Using a Hausdorff measure analogue of the inhomogeneous Khintchine-Groshev Theorem, ... More
Unsupervised Predictive Memory in a Goal-Directed AgentMar 28 2018Animals execute goal-directed behaviours despite the limited range and scope of their sensors. To cope, they explore environments and store memories maintaining estimates of important information that is not presently available. Recently, progress has ... More
DeepMind LabDec 12 2016Dec 13 2016DeepMind Lab is a first-person 3D game platform designed for research and development of general artificial intelligence and machine learning systems. DeepMind Lab can be used to study how autonomous artificial agents may learn complex tasks in large, ... More
Neural Episodic ControlMar 06 2017Deep reinforcement learning methods attain super-human performance in a wide range of environments. Such methods are grossly inefficient, often taking orders of magnitudes more data than humans to achieve reasonable performance. We propose Neural Episodic ... More
Human-level performance in first-person multiplayer games with population-based deep reinforcement learningJul 03 2018Recent progress in artificial intelligence through reinforcement learning (RL) has shown great success on increasingly complex single-agent environments and two-player turn-based games. However, the real-world contains multiple agents, each learning and ... More
A Mass Transference Principle for systems of linear forms and its applicationsMar 29 2017May 15 2017In this paper we establish a general form of the Mass Transference Principle for systems of linear forms conjectured in [1]. We also present a number of applications of this result to problems in Diophantine approximation. These include a general transference ... More
Distribution Of Sequences Generated By Certain Simply-Constructed Normal NumbersNov 04 2015In 1949 Wall showed that $x = 0.d_1d_2d_3 \dots$ is normal if and only if $(0.d_nd_{n+1}d_{n+2} \dots)_n$ is a uniformly distributed sequence. In this article, we consider sequences which are slight variants on this. In particular, we show that certain ... More
Model-Free Episodic ControlJun 14 2016State of the art deep reinforcement learning algorithms take many millions of interactions to attain human-level performance. Humans, on the other hand, can very quickly exploit highly rewarding nuances of an environment upon first discovery. In the brain, ... More
Overcoming catastrophic forgetting in neural networksDec 02 2016Jan 25 2017The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Neural networks are not, in general, capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist ... More
Dynamic Backward Slicing of Rewriting Logic ComputationsMay 13 2011Jun 06 2011Trace slicing is a widely used technique for execution trace analysis that is effectively used in program debugging, analysis and comprehension. In this paper, we present a backward trace slicing technique that can be used for the analysis of Rewriting ... More
Psychlab: A Psychology Laboratory for Deep Reinforcement Learning AgentsJan 24 2018Feb 04 2018Psychlab is a simulated psychology laboratory inside the first-person 3D game world of DeepMind Lab (Beattie et al. 2016). Psychlab enables implementations of classical laboratory psychological experiments so that they work with both human and artificial ... More
Debugging of Web Applications with Web-TLRAug 11 2011Web-TLR is a Web verification engine that is based on the well-established Rewriting Logic--Maude/LTLR tandem for Web system specification and model-checking. In Web-TLR, Web applications are expressed as rewrite theories that can be formally verified ... More
Dust properties of the cometary globule Barnard 207 (LDN 1489)Jul 25 2017Barnard 207 (B207, LDN 1489, LBN 777), also known as the Vulture Head nebula, is a cometary globule in the Taurus-Auriga-Perseus molecular cloud region. B207 is known to host a Class I protostar, IRAS 04016+2610, located at a projected distance of ~8,400 ... More
The Internals of the Data CalculatorAug 06 2018Data structures are critical in any data-driven scenario, but they are notoriously hard to design due to a massive design space and the dependence of performance on workload and hardware which evolve continuously. We present a design engine, the Data ... More
Parallel WaveNet: Fast High-Fidelity Speech SynthesisNov 28 2017The recently-developed WaveNet architecture is the current state of the art in realistic speech synthesis, consistently rated as more natural sounding for many different languages than any previous system. However, because WaveNet relies on sequential ... More
Latent Alignment and Variational AttentionJul 10 2018Nov 07 2018Neural attention has become central to many state-of-the-art models in natural language processing and related domains. Attention networks are an easy-to-train and effective method for softly simulating alignment; however, the approach does not marginalize ... More
Evaluating Compositionality in Sentence EmbeddingsFeb 12 2018May 17 2018An important challenge for human-like AI is compositional semantics. Recent research has attempted to address this by using deep neural networks to learn vector space embeddings of sentences, which then serve as input to other tasks. We present a new ... More
Building Machines that Learn and Think for Themselves: Commentary on Lake et al., Behavioral and Brain Sciences, 2017Nov 22 2017We agree with Lake and colleagues on their list of key ingredients for building humanlike intelligence, including the idea that model-based reasoning is essential. However, we favor an approach that centers on one additional ingredient: autonomy. In particular, ... More