Prof. Gal Chechik

Date: Monday, 12:00-13:00

Title: Personalizing foundation models for visual generative AI 


Machine learning now heavily relies on "foundation models", representations trained with massive data in a task-agnostic way, and then used for downstream tasks. But how can we make them most useful for our own tasks? I will discuss a series of studies addressing this problem in the context of vision-and-language models. These include textual-inversion techniques - where we teach new personalized visual concepts to the model that can be combined with known concepts, and ways to improve their editability and speed. I will also discuss generation of rare classes, personalized approaches to discriminative problems, and future directions in model personalization. 


Prof. Shannon Vallor

Date: Monday, 16:30-17:30

Title: The AI Mirror: Human Prediction and Reflection in AI


Would you ever try to chart your path up a dangerous, unfamiliar mountain while looking in a mirror facing behind you? Today’s AI technologies are marketed as the key to predicting and navigating humanity’s uncertain future in a time of crisis. Yet are these new tools clear windows into our future, or are they looking-glass reflections of our past? Can they ever show us what we and our societies can become, where we might go anew, or what is possible for humanity to accomplish together for the first time? In the face of growing planetary and civilizational challenges that require letting go of the unsustainable ways of the past, humanity’s most urgent task is to embrace and renew our capacities for self-creation, moral imagination and above all, wisdom. AI too has a vital role to play in that task – if we have the courage to reclaim, rethink and rebuild these technologies in the name of humane futures.


Prof. Dr. Ryan Cotterell

Date: Tuesday, 8:30-9:30

Ttitle: Concept Erasure: Finding Interpretable Subspaces in Neural Representations


Abstract. Modern neural models rely on pre-trained representations that emerge without direct supervision. As these representations are increasingly used in real-world applications, failure to understand what the representations encode, paired with the corresponding inability to control their content, is a growing problem. This talk discusses a new paradigm, concept erasure, under which we identify subspaces within the pre-trained representation space that encode a targeted concept. Identifying such concept spaces affords us a better understanding about the content of the representations. For instance, given that most generative language models nearly always generate grammatically correct sentences, e.g., “The children in the park are playing” instead of “The children in the park is playing”, it is reasonable to ask what part of the model’s representations encodes the long-distance subject-verb agreement. In the theoretical portion of the talk, we present a variety of algorithms to perform concept erasure, some with guarantees. In the empirical portion, we show how to use our algorithms to reliably discover human-interpretable subspaces.


Prof. Emma Brunskill

Date: Tuesday, 16:30-17:30

Title: Towards Responsible Reinforcement Learning


Reinforcement learning has had many exciting results in the last decade, from video games to robotics to most recently advancing results for ChatGPT. Many domains, including healthcare and education, involve making sequences of decisions that have impact on people. Such settings deserve responsible reinforcement learning algorithms that are fair, safe and robust. In this talk I’ll discuss some of my and my lab’s work on tackling these challenges in responsible reinforcement learning and multi-armed bandits, and I will provide examples from our applications to education, healthcare and criminal justice. 

Date: Tuesday, 16:30-17:30

Prof. Ariel Procaccia

Date: Wednesday, 9:00 - 10:00  

Title: Democracy and the Pursuit of Randomness


Sortition is a storied paradigm of democracy built on the idea of choosing representatives through lotteries instead of elections. In recent years this idea has found renewed popularity in the form of citizens’ assemblies, which bring together randomly selected people from all walks of life to discuss key questions and deliver policy recommendations. A principled approach to sortition, however, must resolve the tension between two competing requirements: that the demographic composition of citizens’ assemblies reflect the general population and that every person be given a fair chance (literally) to participate. I will describe our work on designing, analyzing and implementing randomized participant selection algorithms that balance these two requirements. I will also discuss practical challenges in sortition based on experience with the adoption and deployment of our open-source system, Panelot.

Best Dissertation Talk


Ulrike Schmidt-Kraepelin

Date: Wednesday, 12:30-13:30

Title: Models and Algorithms for Scalable Collective Decision Making


This PhD thesis derives models and algorithms for a large variety of collective choice problems, while placing its focus on scalability. That is, even though parts of the problem---the number of outcomes, of decisions, or of agents--- may become very large, the goal is to keep the cognitive effort for the agents and the computational effort of determining the outcome of collective choice rules low. Beyond that, the thesis particularly focuses on properties of collective choice rules that capture (proportional) representation of the agents' preferences. In particular, it studies multi-winner elections, weighted fair allocation, liquid democracy, and the elicitation and aggregation of pairwise preferences. 

Honorary patronage


Jacek Majchrowski, Mayor of the City of Kraków


Co-organized and supported by


Jagiellonian University in Kraków


AGH University of Science and Technology

Krakow Technology Park

Silver Sponsors

Bronze Sponsors


Media Partners


Official Community Partner