bumblekite MLSS 2023
schedule in detail
Learn more about our 2023 sessions, including a set of reading materials that are recommended to be read beforehand.
Technical note: Find out more about each session of the particular day below by clicking on the line of the day.
Full schedule
July 2nd, Sun - arrivals, social programme
15:30-18:30 Hike to Uetliberg
17:00-19:00 Guided city tour with Emre
July 3rd, Mon - clinical, intro
9:30 lecture, Judy Gichoya
11:30 leadership conversation, Matthias Egger & Miriam Donaldson
Topic: new financing mechanisms to fund innovation
14:00 tutorial, Christoph Berns
Topic: Common patterns of Data Science use cases: How embeddings can be utilized to solve diverse Data Science challenges in the medical domain
reading list
Medical Coding with Biomedical Transformer Ensembles and Zero/Few-shot Learning, NAACL-HLT 2022
Discovering Key Topics From Short, Real-World Medical Inquiries via Natural Language Processing, Front. Comput. Sci., 22 September 2021 Sec. Digital Public Health
Attention is All you Need, Advances in Neural Information Processing Systems 30 (NIPS 2017)
Hidden Technical Debt in Machine Learning Systems, Neural Information Processing Systems
ProteinBERT: a universal deep-learning model of protein sequence and function, Bioinformatics
Machine Learning Design Patterns, O'Reilly
Agile Data Science 2.0, O'Reilly
18:00 leadership conversation, Miriam Donaldson, Jelena Curcic, Judy Gichoya, Christoph Berns & Sebastiano Caprara
Topic: building data science & machine learning teams
July 4th, Tue - multimodal, sensors data
9:00 lecture, Christian Holz
Title: Zero-effort Mobile Health for Precision Medicine
In this talk, we will walk through the challenges and opportunities for mobile health in the domain of diagnostics to collect and process representative data, captured from patients and everyday consumers "in the wild" and outside controlled scenarios.
11:00 communication workshop, Mirna Šmidt
12:15 tutorial, Julian Lechuga
Topic: Open source in healthcare
We will discuss the advantages and disadvantages of open source projects and datasets in the healthcare ecosystem. A comparison of the JAX and Pytorch frameworks will be performed in the multimodal setting when addressing uncertainty quantification for healthcare.
reading list
- MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification
Hayat, Nasir, Krzysztof J. Geras, and Farah E. Shamout. "MedFuse: Multi-modal fusion with clinical time-series data and chest X-ray images." Machine Learning for Healthcare Conference. PMLR, 2022.
Filos, Angelos, et al. "A systematic comparison of Bayesian deep learning robustness in diabetic retinopathy tasks."(2019).
Rudner, Tim GJ, et al. "Tractable function-space variational inference in Bayesian neural networks." Advances in Neural Information Processing Systems 35 (2022): 22686-22698.
Band, Neil, et al. "Benchmarking bayesian deep learning on diabetic retinopathy detection tasks."(2022).
18:00 leadership conversation, Jennifer Pougnet, Herko Coomans, Gian-Reto Grond
Topic: health data strategy and policy
July 5th, Wed - omics & imaging
9:00 lecture, Jonas Zierer
11:00 communication workshop, Gregor Willis
12:15 tutorial, Judy Gichoya
Title: Decoding the Invisible: A Comprehensive Guide to Understanding and Overcoming Limitations of explanations in Radiology AI Image Interpretation
reading list
18:00 leadership conversation, Iris Shih, Claudia Mazzà, Alf Scotland
19:30 leadership conversation, Antonija Burcul
July 6th, Thu - clinical trials & EHR
9:00 lecture, Lukas Widmer
Title: My journey to Novartis and into Bayesian Modelling & Evidence Synthesis in Oncology
In early clinical trials, data are inherently sparse and expensive to acquire, and big data can be ethically problematic to gather. In early drug development for Oncology, a key objective is to identify a range of safe doses for novel drugs or combination therapies being tested in human cancer patients for the first time. I will give an introduction to this problem space as a case of an inherently small-data problem, Bayesian modelling and evidence synthesis techniques that can be employed, and illustrate how these may be applied in Phase I Oncology dose escalation trials.
reading list
Widmer, L. A., Bean, A., Ohlssen, D., and Weber, S. Principled Drug-Drug Interaction Terms for Bayesian Logistic Regression Models of Drug Safety in Oncology Phase I Combination Trials, arXiv e-prints (2023)