Nikola Mrkšić

Nikola Mrkšić

London, England, United Kingdom
9K followers 500+ connections

Join to follow

About

Co-founder and CEO of PolyAI, a leading supplier of Conversational AI for automated…

Articles by Nikola

Activity

Join now to see all activity

Experience

  • PolyAI Graphic

    PolyAI

    London, United Kingdom

  • -

    Cambridge, United Kingdom

  • -

    Cambridge, United Kingdom

Education

  • University of Cambridge Graphic

    University of Cambridge

    Activities and Societies: Cambridge University Serbian Society

    Research focused on using deep learning techniques to advance the language understanding
    capabilities of spoken dialogue systems.

    Research areas: language understanding, representation learning, word embeddings,
    semantic specialisation, cross-lingual semantics, domain adaptation.

    Supervisor: Professor Steve Young, Dialogue Systems Group, Department of Engineering.

  • Part III of the Computer Science Tripos; Master's degree focused on preparing students for doctoral research.

    Research Project: "Kernel Structure Discovery for Gaussian Process Classification", supervised by Professor Zoubin Ghahramani at the Machine Learning Group.

  • Final year project: "Semi-supervised Learning Methods for Data Augmentation", supervised by Dr Sean Holden at the Computer Laboratory. First Class Honours in the final examinations.

  • -

    Activities and Societies: Physics seminar.

    Currently teaching assistant at the Physics Seminar - lectured on Quantum Computing, Mathematics and Artificial Intelligence.

  • -

    Activities and Societies: President of Parliament, Sep/2009 - Jun/2010; Silver Medal at the Balkan Olympiad in Informatics.

Publications

  • Multi-domain Dialog State Tracking using Recurrent Neural Networks

    Proceedings of the Association for Computational Linguistics (ACL) 2015, Beijing, China

    Dialog state tracking is a key component of many modern dialog systems, most of which are designed with a single, well-defined domain in mind. This paper shows that dialog data drawn from different dialog domains can be used to train a general belief tracking model which can operate across all of these domains, exhibiting superior performance to each of the domain-specific models. We propose a training procedure which uses out-of-domain data to initialise belief tracking models for…

    Dialog state tracking is a key component of many modern dialog systems, most of which are designed with a single, well-defined domain in mind. This paper shows that dialog data drawn from different dialog domains can be used to train a general belief tracking model which can operate across all of these domains, exhibiting superior performance to each of the domain-specific models. We propose a training procedure which uses out-of-domain data to initialise belief tracking models for entirely
    new domains. This procedure leads to improvements in belief tracking performance regardless of the amount of in-domain data available for training the model.

    Other authors
    See publication

Courses

  • Algebraic Path Routing

    -

  • Algorithms

    -

  • Analysis I

    -

  • Artificial Intelligence I

    -

  • Artificial Intelligence II

    -

  • Automated Reasoning

    -

  • Bioinformatics

    -

  • Business Studies

    -

  • C/C++

    -

  • Compiler Construction

    -

  • Complexity Theory

    -

  • Computation Theory

    -

  • Computer Design

    -

  • Computer Graphics & Image processing

    -

  • Computer Networking

    -

  • Computer Systems Modelling

    -

  • Computer Vision

    -

  • Concepts of Programming Languages

    -

  • Concurrent & Distributed Systems

    -

  • Databases

    -

  • Differential Equations

    -

  • Digital Electronics

    -

  • Discrete Mathematics

    -

  • Economics & Law

    -

  • Foundations of CS (ML)

    -

  • Hoare Logic

    -

  • Information Retreival

    -

  • Information Theory

    -

  • Interactive Formal Verification

    -

  • Java & Further Java

    -

  • Lexical Semantics

    -

  • Logic and Proof

    -

  • Machine Learning

    -

  • Mathematical Methods for CS

    -

  • Natural Language Processing

    -

  • Object Oriented Programming

    -

  • Operating systems

    -

  • Principles of Communications

    -

  • Probability

    -

  • Quantum Computing

    -

  • Regular Languages and Finite Automata

    -

  • Security I

    -

  • Semantics of Programming Languages

    -

  • Spoken Language Processing

    -

  • Topics in Concurrency

    -

  • Vectors & Matrices

    -

Projects

  • Semi-supervised Learning Methods for Data Augmentation

    -

    The original goal of this project was to investigate the extent to which data augmentation schemes based on semi-supervised learning algorithms can improve classification accuracy in supervised learning problems. The objectives included determining the appropriate algorithms, customising them for the purposes of this project and providing their Matlab implementations. These algorithms were to be used to develop a robust system for achieving data augmentation in arbitrary application areas. For…

    The original goal of this project was to investigate the extent to which data augmentation schemes based on semi-supervised learning algorithms can improve classification accuracy in supervised learning problems. The objectives included determining the appropriate algorithms, customising them for the purposes of this project and providing their Matlab implementations. These algorithms were to be used to develop a robust system for achieving data augmentation in arbitrary application areas. For evaluation purposes, a general framework for assessing the quality of data augmentation achieved was to be constructed.

    The project met and exceeded all of the success criteria. A survey of theoretical results underlying data augmentation has been conducted. Full, general implementations of Bayesian Sets, Spy-EM and Roc-SVM algorithms, as well as their proposed extensions have been implemented. A general scheme for achieving data augmentation in binary and multi-class classification has been developed and successfully applied to the three application areas proposed. An evaluation framework for assessing the quality of data augmentation was implemented and used to give statistical significance to the results obtained.

    Final Dissertation Mark: 80

    Other creators
    • Dr Sean Holden was my supervisor for this project.

  • Locus - undergraduate research opportunities program 2011

    -

    Created a secure location sharing Android application using the "Nigori" cryptographic protocol. Effectively, this application provides a distributed (i.e. non server-oriented) service for secure location sharing between trusted parties.

    Other creators
    • Dr Alastair Beresford and Dr Andrew Rice were my supervisors for this project.
    See project

  • Distribution of wealth in the society

    -

    Econophysics research trying to extend some of the existing methods used in the field to obtain new information about macroeconomic factors' effects on the distribution of wealth.

    Other creators

Honors & Awards

  • Research Scholar (Trinity College, Cambridge)

    Trinity College, Cambridge

  • Senior Scholar (Trinity College, Cambridge)

    Trinity College, Cambridge

  • Silver Medal

    Balkan Olympiad in Informatics

Languages

  • English

    Native or bilingual proficiency

  • Serbian

    Native or bilingual proficiency

  • French

    Limited working proficiency

More activity by Nikola

View Nikola’s full profile

  • See who you know in common
  • Get introduced
  • Contact Nikola directly
Join to view full profile

Other similar profiles

Explore collaborative articles

We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.

Explore More

Others named Nikola Mrkšić

3 others named Nikola Mrkšić are on LinkedIn

See others named Nikola Mrkšić

Add new skills with these courses

See all courses