The main idea of the project is to explore opportunities for applying continuous time-forecasting algorithms (like Neural ODEs) for long-horizon time-series prediction, including a faster generative visual AI inference (like the stable diffusion).

We create new solutions to strengthen private, public and state security. We develop, among others, multi-level management systems to protect critical infrastructure as well as systems for securing key state services against both kinetic and cyber threats. Our algorithms develop concepts successfully used in the United States to improve the security of airports, sea terminals, transmission infrastructure, or to enhance the protection of endangered animal species in national parks in many places around the world. Furthermore, we develop AI-based solutions that increase the effectiveness of unmanned platforms in security-related applications.

I am interested in methods that can deliver a robust decision-making capability in complex scenarios. Such methods can be applied in control to obtain broadly intelligent agents.

At the core of this effort is solving credit assignment over prolonged horizons. I like to put, perhaps unorthodoxly, many problems under this umbrella. The tasks requiring extreme reasoning, like math solving, on the one side and classical continuous control in robotics, on the other side.

I will present research topics in several domains of the modern AI reserach:

1. sequential modeling and transformers
2. reinforcement learning
3. multi-task and continual learning
4. planning and learning

In this talk I am going to briefly describe the topic of my upcoming ERC grant. The project is set in the field of computational social choice. Its main goal is to analyse formal models describing scenarios, where a group of individuals disagrees on certain matters, yet needs to make a collective decision. This can be observed, for example, in the elections of representative bodies or in participatory budgeting.

Usually, automata have yes/no outputs, i.e. they define languages. There is also a theory about automata with more complicated outputs, e.g. automata that define string-to-string functions. These automata can be quite powerful, looking like Turing complete programs to the untrained eye, and yet their theory retains much of the beauty and simplicity of the theory of finite automata.

Concurrent systems are ubiquitous, and the impact of their reliability is continuously increasing. In consequence, we observe an increasing call for improving the quality of concurrent systems. It is well known that concurrent system designs are prone to subtle bugs that are inherently difficult to find by humans, and examples of costly damages caused by such bugs abound. One way of achieving improvement of quality of such systems is formal verification, i.e., automatic analysis methods.

The topic of the PhD is to push further theoretical foundations of automatic analysis of models of concurrent systems, by investigating the frontiers of such analysis, and to enhance its practical applicability. We will specifically concentrate on the model of Petri nets, also known as vector addition systems, a long established model of concurrency with extensive applications in modelling and analysis of hardware, software and database systems, as well as chemical, biological and business processes. The central algorithmic problem for Petri nets is reachability: whether from the given initial configuration there exists a sequence of valid execution steps that reaches the given final configuration. The complexity of the problem has remained unsettled for over 40 years, becoming a most important open question in formal models of concurrent systerms, until the recent series of breakthrough results by Czerwiński, Lasota, Lazic, Leroux, Mazowiecki and Orlikowski (2019-2021) that closed the problem. On the technical level, the topic of the PhD is to exploit novel ideas underlying these results, in order to identify tractable special cases, variants and extensions of the problem.

We will outline the profile of the research group gathered around project BOBR: Decomposition Methods for Discrete Structures. The main theme here is to study a variety of decomposition concepts (e.g. tree decompositions, topological embeddings, low-complexity coverings, logical interpretations,...) offered by structural graph theory and finite model theory, with the ultimate goal of providing new techniques and tools for the design of efficient algorithms on well-structured inputs (mostly graphs). On the combinatorial side, we mostly rely on advances of the graph minors theory and the theory of sparsity, and our goal is to build a sound theory of well-structured dense graphs grounded on concepts inspired by model theory (particularly, stability theory). On the algorithmic side, the developed tools are mostly applicable in the paradigms of parameterized algorithms and approximation algorithms, and we often consider computational problems coming from finite model theory (for instance, model-checking).

In the first part of my talk, I will briefly present how IDEAS NCBR PhD programme works (https://ideas-ncbr.pl/en/doctoral-schools/). In the second part, I will introduce research groups and research topics present at IDEAS NCBR.

In this talk I present a series of related topics in the mathematical modelling and computational methods for molecular medicine.

Text indexing is about preprocessing a given text into a data structure that allows answering queries about occurrences of specified patterns in the text quickly. This classic problem has been addressed with much success, starting from the classical data structures like the suffix tree and array, and continuing with more modern, space efficient solutions. We will consider variants of the text indexing problem that are motivated by practice (for example, indexing with few errors), current trends in theory (indexing labeled graphs) and basic research questions (computing string statistics). We aim at worst-case efficient data structures, breaking theoretical barriers and, possibly, conditional hardness proofs. One 4-year PhD scholarship is offered in the project starting in Fall 2023.

For fixed integer t, the class of Pt-free graphs consists of graph without the t-vertex path as an induced subgraph. It is conjectured that a number of combinatorial problems, including Maximum Independent Set, remains polynomial-time solvable in Pt-free graphs (for every fixed t). In the recent years, we have seen a number of results supporting this conjecture, but we are still far from understanding the structure of such graphs to fully prove it.

Resolving the conjecture is one of the first main goals of my NCN SONATA BIS project that is about to start, and where I have positions for PhD students.

In the talk I will highlight the main recent results, proof techniques, and possible next steps.

I will be talking about some of the open problems in the area of reinforcement learning and language models. Then I will discuss several related interesting future research directions, including open-endedness, meta-learning of algorithms, automatic generation of environments and rewards, truthfulness, and alignment.

I will present a brief overview of the research conducted by my group at the University of Warsaw and IDEAS NCBR, spanning from theory to applications, with a particular focus on blockchain technology. I will also highlight potential Ph.D. topics in this field.

Our lab, called Computational Medicine, focuses on machine learning for medicine applications. Specifically, we work with probabilistic graphical models, as well as deep generative models, with a recent focus on variational auto-encoders. In my talk, I will briefly cover our AI approaches to oncology, antimicrobial peptide generation, and covid19.