Embedded Files
Session 1
09:00-09:10 Opening remarks
Organizing Committee
09:10-09:30 Introduction to Epistemic AI
Organizing Committee
09:30-10:00 Invited talk* by Gert De Cooman
10:00-10:30 Coffee break
Session 2
10:30-11:30 Best student paper and best paper award oral presentation
URL: A Representation Learning Benchmark for Transferable Uncertainty Estimates
Michael Kirchhof (University of Tübingen)*; Bálint Mucsányi (University of Tübingen); Seong Joon Oh (Naver AI Lab); Enkelejda Kasneci (University of Tuebingen)
Joint Modelling for Uncertainty Quantification
Chunlin Ji (Kuang-Chi)*; Dingwei Gong (Sun Yat-sen University); Meiying Zhang (SUSTech)
11:30-12:30 Poster session #1
Massively Parallel Reweighted Wake-Sleep
Thomas E Heap (University of Bristol)*; Gavin Leech (University of Bristol); Laurence Aitchison (University of Bristol)
EnSolver: Uncertainty-Aware CAPTCHA Solver Using Deep Ensembles
Duc C Hoang (Florida International University)*; Cuong V Nguyen (Florida International University); Amin Kharraz (Florida International University
Defensive Perception: Estimation and Monitoring of Neural Networks Performance under Deployment
Hendrik Vogt (ZF Friedrichshafen AG)*; Mark Schutera (ZF Friedrichshafen AG); Stefan Bühler (ZF)
Offline Reinforcement Learning with Pessimistic Value Priors
Filippo Valdettaro (Imperial College London)*; Aldo Faisal (Imperial College London)
Semantic Attribution For Explainable Uncertainty Quantification
Hanjing Wang (Renselaer Polytechnic Institute)*; Dhiraj Joshi (IBM Research AI); Shiqiang Wang (IBM Research); Qiang Ji (Renselaer Polytechnic Institute)
12:30-13:00 Causal inference meets epistemic AI: practical, general, methods for counterfactuals and information fusionInvited talk
The field of causal inference is naturally suited to meet the paradigm of sets of models (or distributions) due the inherent `non-identifiability' of its models: very often there are many causal models that give rise to the same distribution in the observed data. Most research efforts have been focused on the identifiable case, though, where causal queries are eventually transformed into probabilistic expressions that can precisely be estimated from data. In contrast, this talk will concentrate on recent advances in the broader setting of partial identification, and in particular on research done at IDSIA. The starting point will be the so-called `expectation-maximisation for causal computations’ (EMCC) scheme that permits quantifying the uncertainty of latent nodes in structural causal models from data about the manifest variables; and from this to bound the result of general counterfactual queries. We shall go on to describe extensions of the EMCC that enable it to deal with selection bias as well as with quite a general paradigm of information fusion, which mixes observational and interventional data from any number of sources. We shall finally touch on a recent connection with probabilistic circuits that allows EMCC-based algorithms to run one order of magnitude faster that it was possible before.
13:00-14:00 Lunch break
Session 3
14:00-14:30 Spotlight (oral) presentation # 1
MiDAS: Domain Learning for Robust Fake News Detection in Distribution Shifts
Abhijit Suprem (Georgia Institute of Technology)*; Calton Pu (Georgia Tech)
Model Robustness and Active Learning with Missing-Not-At-Random Outcomes
Alan Mishler (J.P. Morgan Chase)*; Mohsen Ghassemi (JP Morgan Chase); Alec Koppel (JP Morgan Chase); Sumitra Ganesh (JPMorgan)
14:30-15:00 Invited talk* by Yarin Gal
Foundation Models That Can Tell Us When They Don’t Know
15:00-16:00 Poster session #2
Deep learning and MCMC with aggVAE for shifting administrative boundaries: mapping malaria prevalence in Kenya
Elizaveta Semenova (University of Oxford)*; Swapnil Mishra (Imperial College London); Samir Bhatt (Imperial College London); Seth Flaxman (Oxford); H Juliette T Unwin (Imperial College London)
Bag of Policies for Distributional Deep Exploration
Asen Nachkov (Imperial College London); Luchen Li (Imperial College London); Giulia Luise (Imperial College London); Filippo Valdettaro (Imperial College London); Aldo Faisal (Imperial College London)*
Towards Understanding the Interplay of Generative Artificial Intelligence and the Internet
Gonzalo Martínez (University Carlos III of Madrid)*; Lauren Watson (The University of Edinburgh); Pedro Reviriego (Reviriego); Jose Alberto Hernandez (Universidad Carlos III de Madrid); Marc Juarez (The University of Edinburgh); Rik Sarkar (The University of Edinburgh)
Nadaraya-Watson Selective Regression
Fedor Noskov (Skoltech); Alexander Fishkov (Skoltech); Maxim Panov (Technology Innovation Institute)
Optimizing Brain Tumor Classification: A Comprehensive Study on Transfer Learning and Imbalance Handling in Deep Learning Models
Raza Imam (Aligarh Muslim University)*; Mohammed Talha Alam (Jamia Hamdard University)
16:00-16:30 Coffee break
Session 4
16:30-17:00 Spotlight (oral) presentation # 2
Fair Ranking under Disparate Uncertainty
Richa Rastogi (Cornell)*; Thorsten Joachims (Cornell)
A Novel Bayes’ Theorem for Upper Probabilities
Michele Caprio (University of Pennsylvania)*; Yusuf Sale (University of Munich (LMU)); Eyke Hüllermeier (University of Munich); Insup Lee (University of Pennsylvania)
17:00-17:30 Distribution-free uncertainty quantification under distribution drifts and shifts
In this talk, Aaditya Ramdas will briefly introduce two recent lines of work for quantifying the uncertainty of the predictions of black-box machine learning algorithms in the presence of distribution drifts and shifts (DDS). The first is "Conformal prediction beyond exchangeability" (Annals of Statistics'23), which sequentially recalibrates conformal scores to produce prediction sets with coverage guarantees under DDS. The second is "Online Platt scaling with Calibeating" (ICML'23), which sequentially recalibrates probabilistic outputs from classifiers in order to remain calibrated under DDS.
17:30-18:00 Panel on the future of uncertainty-aware models and their applications
18:00-18:30 Award ceremony and closing remarks
Organizing Committee
*: The titles of the invited talks will be announced soon.
Page updated
Google Sites
Report abuse