Egor Kraev

Egor Kraev

Zürich, Zürich, Schweiz
3779 Follower:innen 500+ Kontakte

Mitglied werden und folgen

Info

With a foundation in the Russian mathematical tradition, refined at at ETH Zurich and the…

Aktivitäten

Anmelden, um alle Aktivitäten zu sehen

Berufserfahrung

  • Motleycrew.ai Grafik

    Motleycrew.ai

    Zurich, Switzerland

  • -

  • -

  • -

    London, United Kingdom

  • -

    London, United Kingdom

  • -

    London, United Kingdom

  • -

    London, United Kingdom

  • -

    London, UK

  • -

  • -

  • -

  • -

  • -

  • -

    Washington D.C. Metro Area

Ausbildung

  • University of Maryland Grafik

    University of Maryland

    –

    Specialization: Compilation, econometric analysis and modeling of macroeconomic and sectoral data
    GPA 4.0 out of 4.0
    Supported by the Swiss Foundation for Gifted Young People

  • –

    Upon graduation received awards for Best Thesis and Best GPA
    GPA 6.0 out of 6.0

Bescheinigungen und Zertifikate

Veröffentlichungen

  • Probabilistic hypergraph grammars for efficient molecular optimization

    arxiv.org

    Veröffentlichung anzeigen

  • Grammars and reinforcement learning for molecule optimization

    arxiv.org

    Veröffentlichung anzeigen

  • Assessing Modeling Approaches to the Distributional Effects of Macroeconomic Reform

    Development Policy Review, Volume 23 Issue 3 Page 299

    Andere Autor:innen

  • The first observed stellar X-ray flare oscillation: Constraints on the flare loop length and the magnetic field

    Astronomy and Astrophysics, 436, 1041-1047 (2005)

    Andere Autor:innen
    • U. Mitra-Kraev
    • L. K. Harra
    • D.R. Williams

  • Accounting Matrix and Transaction Matrices: A Concise Formalism for Describing Financial Stock Dynamics

    Working Paper, CEPA, New School for Social Research, New York

  • Structural Adjustment Policies in Ghana in the 1990s: An Empirical Analysis and Policy Recommendations

    UNDP Discussion Paper and Policy Summary Paper

  • Modeling Macroeconomic and Distributional Impacts of Stabilization and Adjustment Packages: Current Literature and Challenges

    Working Paper 2003-6, CEPA, New School for Social Research, New York

  • Ceteris Non Paribus: Integrating Ecological, Social and Economic Systems in a Dynamic Model of Island Tourism

    Ecological Modeling, 175 (2004) 121-136

    Andere Autor:innen
    • T.Patterson
    • T.Gulden
    • K.Cousins

  • Stocks, Flows and Complementarity: Formalizing a Basic Insight of Ecological Economics - Ecological Economics

    Ecological Economics, 43 (2002) 277-286

  • Existence and Uniqueness for Height Structured Hierarchical Population Models

    Natural Resource Modeling, vol.14 no.1 pp.45-70, Spring 2001

Patente

  • Infant development system

    GB GB2554344

Projekte

  • Wise-Pizza https://github.com/transferwise/wise-pizza

    Go beyond averages, to a concise description of what segments of your dataset are different from the rest, for example what customer groups drove the change in growth rates from one quarter to the next.

    Andere Mitarbeiter:innen

  • CausalTune https://github.com/py-why/causaltune

    The first AutoML package for causal inference models, allowing you to select the best uplift model for your A/B test, and not just to calculate the average impact of your test, but to identify user groups whose response was stronger or weaker than the rest.

  • Molecular optimization using reinforcement learning

    –Heute

    Generate molecules maximizing given benchmarks

    Techniques used: multiple reinforcement learning flavors: batch-advantage policy gradient, model-guided Monte Carlo Tree Search, memetic algorithms (genetic algorithms at model level with local optimization in between)

    Ongoing work: Meta-parameter optimization for a memetic algorithm, on the Guacamol benchmarks

    Papers:

    Probabilistic hypergraph grammars for efficient molecular optimization, E.Kraev and M.Harley, May…

    Generate molecules maximizing given benchmarks

    Techniques used: multiple reinforcement learning flavors: batch-advantage policy gradient, model-guided Monte Carlo Tree Search, memetic algorithms (genetic algorithms at model level with local optimization in between)

    Ongoing work: Meta-parameter optimization for a memetic algorithm, on the Guacamol benchmarks

    Papers:

    Probabilistic hypergraph grammars for efficient molecular optimization, E.Kraev and M.Harley, May 2019, https://arxiv.org/abs/1906.01845: Infer a hypergraph grammar from the ChEMBL dataset; using empirical conditional probabilities as priors for next rule selection, and batch advantage policy gradient applied to a trivial model, achieve competitive performance on the 20 advanced GuacaMol benchmarks, with only hundreds of training steps from a cold start

    Grammars and reinforcement learning for molecule optimization, E.Kraev, Nov 2018, https://arxiv.org/abs/1811.11222: Hand-crafted context-free SMILES grammar, Transformer encoder-decoder outputting sequence of grammar rules and best-of-batch policy gradient to achieve then state-of-the-art penalized logP scores

    Projekt anzeigen

  • Multi-language grammar tree generation

    –

    Using the universaldependencies.org dataset and Polyglot embeddings, cast grammar tree generation for natural language sentences as a classification problem. Modifying the Transformer architecture by feeding the attention matrix and its transpose as additional inputs into each Transformer layer was shown to substantially improve performance.

    Projekt anzeigen

Sprachen

  • English

    Muttersprache oder zweisprachig

  • German

    Muttersprache oder zweisprachig

  • Russian

    Muttersprache oder zweisprachig

  • French

    Gute Kenntnisse

Erhaltene Empfehlungen

8 Personen haben Egor Kraev empfohlen

Jetzt anmelden und ansehen

Weitere Aktivitäten von Egor Kraev

Egor Kraevs vollständiges Profil ansehen

  • Herausfinden, welche gemeinsamen Kontakte Sie haben
  • Sich vorstellen lassen
  • Egor Kraev direkt kontaktieren
Mitglied werden. um das vollständige Profil zu sehen

Weitere ähnliche Profile