• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site


Mini-Workshop: Stochastic Processes and Probabilistic Models in Machine Learning

Event ended

On September 12 and 13 the laboratory invites four international researchers to give talks on modern applications of stochastic processes and probabilistic modeling in machine learning. The speakers will give an overview of the theory of Dirichlet, Pitman-Yor, Gamma and Gaussian processes, and how they can be applied to deal with large scale problems, interpretability and other tasks. Also, several speakers from Russian scientific groups will present their research.

Language: English

Location: Faculty of Computer Science, Moscow, Kochnovsky Proezd, 3, room 205

Dates: September, 12 and 13.

Materials: slides.

September, 12, room 205

14:00-15:30  Ethical Machine Learning

Novi Quadrianto

Assistant Professor, University of Sussex, Great Britain
Scientific Advisor, HSE Laboratory of Deep Learning and Bayesian Methods

Machine learning technologies have permeated everyday life and it is common nowadays that an automated system makes decisions for/about us, for example, perhaps deciding who is going to get a VTB 24 bank loan. Addressing ethical and legal aspects posed by those technologies constitutes a pressing problem. The long-term goal of our research group is to develop a machine learning framework with plug-and-play ethical and legal constraints that is able to handle fairness, confidentiality, and transparency constraints, their combinations, and also new constraints that might be stipulated in the future. In this talk, I will be mostly discussing background research in fairness and transparency, before presenting our own work in unifying several notions of fairness.

15:45-17:15 Implicit generative models: dual and primal approaches

Iliya Tolstikhin

Postdoc, Max Planck Institute for Intelligent Systems, Tübingen, Germany

The fields of unsupervised generative modelling and representation learning are rapidly growing. Empirical success of recently introduced methods, including  Variational Auto-Encoders (VAE) and Generative Adversarial Nets (GAN), attracts attention many researchers working in various areas of Machine Learning. Last few years led to unprecedented amount of papers, trying to improve the performance of VAEs/GANs, introducing new versions of these algorithms, and coming up with completely new ideas. In this talk I will try to present a unifying view on many of the existing methods, showing that VAEs/GANs are approaching very similar objectives --- f-divergences, integral probability metrics, optimal transports --- from their primal/dual formulations respectively. I will discuss certain consequences of this duality and mention a recent work on optimal transport, establishing interesting links between VAEs/GANs.

17:30-19:00  Introduction to Dirichlet Processes and their use

Wray Buntine

Professor, Monash University, Melbourne, Australia

Assuming the attendee has knowledge of the Poisson, Gamma, multinomial and Dirichlet distributions, this talk will present the basic ideas and theory to understand and use the Dirichlet process and its close relatives, the Pitman-Yor process and the gamma process.  We will first look at some motivating examples.  Then we will look at the non-hierarchical versions of the processes, which are basically infinite parameter vectors.  These have a number of handy properties and have simple, elegant marginal and posterior inference.  Finally, we will look at the hierarchical versions of these processes.  These are fundamentally different.  To understand the hierarchical version we will briefly review some aspects of stochastic process theory and additive distributions.  The hierarchical versions becomes Dirichlet and Gamma distributions (the process part disappears) but the techniques developed for the non-hierarchical process models can be borrowed to develop good algorithms, since the Dirichlet and Gamma are challenging when placed hierarchically.

September, 13, room 205

14:00-15:30  Gaussian Processes

Maurizio Filippone

Assistant Professor, EURECOM, France

The study of complex phenomena through the analysis of data often requires us to make assumptions about the underlying dynamics. In modern applications, for many systems of interest we are facing the challenge of doing so when very little is known about their mechanistic description. Even when a mechanistic description is available, simulating such systems is so computationally expensive that we cannot use it effectively. While probabilistic models based on Gaussian Processes (GPs) offer attractive tools to tackle these challenges in a principled way and to allow for a sound quantification of uncertainty, carrying out inference for these models poses huge computational challenges that arguably hinder their wide adoption. Recent contributions, however, make it possible to massively improve scalability of GPs and to exploit parallel and distributed computing, so that GPs are now in the position to compete with Deep Neural Networks on various problems where the latter are usually the preferred choice. This lecture aims to expose students to the fundamentals of modern GP research. In particular, the lecture will touch upon these points: (i) the basics of GPs; (ii) the challenges in employing GPs for large scale learning problems; (iii) recent advances in GP research to allow for scalable and accurate quantification of uncertainty; (iv) interesting applications of GPs; (v) current and future directions.

15:45-18:30 Russian researchers session
Yuriy Kuratov, Idris YusupovSkill-based Conversational Agent
Mikhail ArkhipovApplication of modern neural architectures to the problem of Russian Named Entity Recognition
Konstantin VorontsovAdditive Regularization for Topic Modeling
Anna PotapenkoInterpretable probabilistic embeddings: bridging the gap between topic models and neural networks
Alexey UmnovData Anonymization with Wasserstein Distance
Viktor YanushLearnable optimization strategies using recurrent neural networks
Oleg IvanovMissing Features Imputation using Conditional Variational Autoencoders
Artyom GadetskyConditional Generators of Words Definitions
Valentin SytovVariational Autoencoders for Image Retrieval