I. Babuschkin
A5085244021- Particle physics theoretical and experimental studies
- Quantum Chromodynamics and Particle Interactions
- High-Energy Particle Collisions Research
- Neutrino Physics Research
- Computational Physics and Python Applications
- Black Holes and Theoretical Physics
- Atomic and Subatomic Physics Research
- Particle Accelerators and Free-Electron Lasers
- Topic Modeling
- Dark Matter and Cosmic Phenomena
- Superconducting Materials and Applications
- Particle Detector Development and Performance
- Scientific Computing and Data Management
- Medical Imaging Techniques and Applications
- Music and Audio Processing
- Software Engineering Research
- Cell Image Analysis Techniques
- Natural Language Processing Techniques
- Generative Adversarial Networks and Image Synthesis
- Stochastic processes and statistical mechanics
- Speech and Audio Processing
- Speech Recognition and Synthesis
- Adversarial Robustness in Machine Learning
- Data Visualization and Analytics
- Reinforcement Learning in Robotics
DeepMind (United Kingdom)
2019-2022
University of Manchester
2016-2018
Istituto Nazionale di Fisica Nucleare, Sezione di Ferrara
2017
Centro Brasileiro de Pesquisas Físicas
2016-2017
Manchester University
2017
We introduce Codex, a GPT language model fine-tuned on publicly available code from GitHub, and study its Python code-writing capabilities. A distinct production version of Codex powers GitHub Copilot. On HumanEval, new evaluation set we release to measure functional correctness for synthesizing programs docstrings, our solves 28.8% the problems, while GPT-3 0% GPT-J 11.4%. Furthermore, find that repeated sampling is surprisingly effective strategy producing working solutions difficult...
Programming is a powerful and ubiquitous problem-solving tool. Systems that can assist programmers or even generate programs themselves could make programming more productive accessible. Recent transformer-based neural network models show impressive code generation abilities yet still perform poorly on complex tasks requiring skills, such as competitive problems. Here, we introduce AlphaCode, system for achieved an average ranking in the top 54.3% simulated evaluations recent competitions...
The recently-developed WaveNet architecture is the current state of art in realistic speech synthesis, consistently rated as more natural sounding for many different languages than any previous system. However, because relies on sequential generation one audio sample at a time, it poorly suited to today's massively parallel computers, and therefore hard deploy real-time production setting. This paper introduces Probability Density Distillation, new method training feed-forward network from...
The ratio of branching fractions ${\cal{R}}(D^{*-})\equiv {\cal{B}}(B^0 \to D^{*-} τ^+ ν_τ)/{\cal{B}}(B^0 μ^+ν_μ)$ is measured using a data sample proton-proton collisions collected with the LHCb detector at center-of-mass energies 7 and 8 TeV, corresponding to an integrated luminosity 3$~$fb$^{-1}$. $τ$ lepton reconstructed three charged pions in final state. A novel method used that exploits different vertex topologies signal backgrounds isolate samples semitauonic decays $b$ hadrons high...
The $\Xi_c^+ K^-$ mass spectrum is studied with a sample of $pp$ collision data corresponding to an integrated luminosity 3.3 fb$^{-1}$, collected by the LHCb experiment. $\Xi_c^+$ reconstructed in decay mode $p K^- \pi^+$. Five new, narrow excited $\Omega_c^0$ states are observed: $\Omega_c(3000)^0$, $\Omega_c(3050)^0$, $\Omega_c(3066)^0$, $\Omega_c(3090)^0$, and $\Omega_c(3119)^0$. Measurements their masses widths reported.
Language modelling provides a step towards intelligent communication systems by harnessing large repositories of written human knowledge to better predict and understand the world. In this paper, we present an analysis Transformer-based language model performance across wide range scales -- from models with tens millions parameters up 280 billion parameter called Gopher. These are evaluated on 152 diverse tasks, achieving state-of-the-art majority. Gains scale largest in areas such as...
The first full amplitude analysis of $B^+\to J/ψϕK^+$ with $J/ψ\toμ^+μ^-$, $ϕ\to K^+K^-$ decays is performed a data sample 3 fb$^{-1}$ $pp$ collision collected at $\sqrt{s}=7$ and $8$ TeV the LHCb detector. cannot be described by model that contains only excited kaon states decaying into $ϕK^+$, four $J/ψϕ$ structures are observed, each significance over $5$ standard deviations. quantum numbers these determined least $4$ lightest has mass consistent with, but width much larger than, previous...
We introduce an approach for deep reinforcement learning (RL) that improves upon the efficiency, generalization capacity, and interpretability of conventional approaches through structured perception relational reasoning. It uses self-attention to iteratively reason about relations between entities in a scene guide model-free policy. Our results show novel navigation planning task called Box-World, our agent finds interpretable solutions improve baselines terms sample complexity, ability...
The first full amplitude analysis of $B^+\to J/ψϕK^+$ with $J/ψ\toμ^+μ^-$, $ϕ\to K^+K^-$ decays is performed a data sample 3 fb$^{-1}$ $pp$ collision collected at $\sqrt{s}=7$ and $8$ TeV the LHCb detector. cannot be described by model that contains only excited kaon states decaying into $ϕK^+$, four $J/ψϕ$ structures are observed, each significance over $5$ standard deviations. quantum numbers these determined least $4$ lightest has mass consistent with, but width much larger than, previous...
Advances in deep generative networks have led to impressive results recent years. Nevertheless, such models can often waste their capacity on the minutiae of datasets, presumably due weak inductive biases decoders. This is where graphics engines may come handy since they abstract away low-level details and represent images as high-level programs. Current methods that combine learning renderers are limited by hand-crafted likelihood or distance functions, a need for large amounts supervision,...
The Dalitz plot analysis technique is used to study the resonant substructures of ${B}^{\ensuremath{-}}\ensuremath{\rightarrow}{D}^{+}{\ensuremath{\pi}}^{\ensuremath{-}}{\ensuremath{\pi}}^{\ensuremath{-}}$ decays in a data sample corresponding $3.0\text{ }\text{ }{\mathrm{fb}}^{\ensuremath{-}1}$ $pp$ collision recorded by LHCb experiment during 2011 and 2012. A model-independent angular moments demonstrates presence resonances with spins 1, 2 3 at high...
A search for the rare decays $B_s^0\to\tau^+\tau^-$ and $B^0\to\tau^+\tau^-$ is performed using proton--proton collision data collected with LHCb detector. The sample corresponds to an integrated luminosity of 3fb$^{-1}$ in 2011 2012. $\tau$ leptons are reconstructed through decay $\tau^-\to\pi^-\pi^+\pi^-\nu_{\tau}$. Assuming no contribution from decays, upper limit set on branching fraction $\mathcal{B}(B_s^0\to\tau^+\tau^-) < 6.8\times 10^{-3}$ at 95% confidence level. If instead assumed,...
Measurements of the cross section for producing b quarks in reaction pp→b¯bX are reported 7 and 13 TeV collisions at LHC as a function pseudorapidity η range 2<η<5 covered by acceptance LHCb experiment. The measurements done using semileptonic decays b-flavored hadrons decaying into ground-state charmed hadron association with muon. sections 72.0±0.3±6.8 154.3±1.5±14.3 μb TeV. ratio is 2.14±0.02±0.13, where quoted uncertainties statistical systematic, respectively. agreement theoretical...
In this paper we propose to study generalization of neural networks on small algorithmically generated datasets. setting, questions about data efficiency, memorization, generalization, and speed learning can be studied in great detail. some situations show that learn through a process "grokking" pattern the data, improving performance from random chance level perfect improvement happen well past point overfitting. We also as function dataset size find smaller datasets require increasing...
The production cross-section of J/ψ pairs is measured using a data sample pp collisions collected by the LHCb experiment at centre-of-mass energy $$ \sqrt{s}=13 TeV, corresponding to an integrated luminosity 279 ±11 pb−1. measurement performed for mesons with transverse momentum less than 10 GeV/c in rapidity range 2.0 < y 4.5. be 15.2 ± 1.0 0.9 nb. first uncertainty statistical, and second systematic. differential cross-sections as functions several kinematic variables pair are compared...
The production of J/$ψ$ mesons is studied in proton-lead collisions at the centre-of-mass energy per nucleon pair $\sqrt{s_{\text{NN}}}=8.16$ TeV with LHCb detector LHC. double differential cross-sections prompt and nonprompt are measured as functions transverse momentum rapidity nucleon-nucleon frame. Forward-to-backward ratios nuclear modification factors determined. results compared theoretical calculations based on collinear factorisation using parton distribution functions, colour glass...
The production of J/ψ mesons in jets is studied the forward region proton-proton collisions using data collected with LHCb detector at a center-of-mass energy 13 TeV. fraction jet transverse momentum carried by meson, z(J/ψ)≡pT(J/ψ)/pT(jet), measured pT(jet)>20 GeV pseudorapidity range 2.5<η(jet)<4.0. observed z(J/ψ) distribution for produced b-hadron decays consistent expectations. However, results prompt do not agree predictions based on fixed-order nonrelativistic QCD. This first...
Production cross-sections of prompt charm mesons are measured using data from $pp$ collisions at the LHC a centre-of-mass energy $5\,$TeV. The sample corresponds to an integrated luminosity $8.60\pm0.33\,$pb$^{-1}$ collected by LHCb experiment. production $D^0$, $D^+$, $D_s^+$, and $D^{*+}$ in bins meson transverse momentum, $p_{\text{T}}$, rapidity, $y$. They cover rapidity range $2.0<y<4.5$ momentum ranges $0 < p_{\text{T}} 10\, \text{GeV}/c$ for $D^0$ $D^+$ $1 $D_s^+$ mesons. inclusive...
A measurement of the cross-section for W → eν production in pp collisions is presented using data corresponding to an integrated luminosity 2 fb−1 collected by LHCb experiment at a centre-of-mass energy $$ \sqrt{s}=8 TeV. The electrons are required have more than 20 GeV transverse momentum and lie between 2.00 4.25 pseudorapidity. inclusive cross-sections, where decays eν, measured be {\sigma}_{W^{+}\to {e}^{+}{\nu}_e}=1124.4\pm 2.1\pm 21.5\pm 11.2\pm 13.0\kern0.5em \mathrm{p}\mathrm{b},...
The first observation of the decays Λb0→χc1pK− and Λb0→χc2pK− is reported using a data sample corresponding to an integrated luminosity 3.0 fb−1, collected by LHCb experiment in pp collisions at center-of-mass energies 7 8 TeV. following ratios branching fractions are measured: B(Λb0→χc1pK−)B(Λb0→J/ψpK−)=0.242±0.014±0.013±0.009,B(Λb0→χc2pK−)B(Λb0→J/ψpK−)=0.248±0.020±0.014±0.009,B(Λb0→χc2pK−)B(Λb0→χc1pK−)=1.02±0.10±0.02±0.05,where uncertainty statistical, second systematic, third due on...