Seminar | TransferLab — appliedAI Institute

Seminar

The TransferLab Seminar is a platform where researchers and engineers share and discuss recent advances in AI/ML, striking a balance between accessibility and mathematical depth.

We organize comprehensive talks on ML, RL, DL and statistics, emphasizing both theoretical understanding and practical implementation. These range from introductory overviews to recent developments - and are open to all interested individuals!

Each talk, lasting about 45min followed by around 15min discussion, focuses on a specific topic related to our ongoing series topics. To stay informed, follow the appliedAI Institute on LinkedIn. If you’re interested in presenting your work, contact us. Recordings of past seminars are available on YouTube.

Upcoming Talks

Nothing here yet!

Previous Talks

Robustly representing uncertainty through sampling in deep neural networks - Fabio Peruzzo (appliedAI Initiative)

Dec 5, 2022

In the previous instalments of our seminar series we have explored how bayesian methods can be used to estimate the epistemic uncertainty of a model, i.e. the uncertainty coming from the choice of network’s parameters. In particular, two weeks ago we have seen how variational inference applied to Bernoulli and Gaussian dropout can make Bayesian NNs tractable under certain assumptions.

Uncertainty Quantification

Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning - Iván Rodríguez (appliedAI Initiative)

Nov 17, 2022

In this presentation, we revisit landmark research that interprets the dropout training process in deep neural networks as approximations of Bayesian inference within deep Gaussian processes. As a direct outcome of this theoretical viewpoint, we gain access to techniques that allow us to model uncertainty within dropout neural networks - essentially salvaging information previously discarded from our existing models.

Uncertainty Quantification

Aleatoric and epistemic uncertainty in machine learning - Michael Panchenko (appliedAI Initiative)

Nov 2, 2022

This talk introduces the notions of aleatoric and epistemic uncertainty for probabilistic models and gives an overview of recent developments in this area. It is mainly based on the corresponding review paper. Aleatoric uncertainty arises from inherent randomness or variability in the data, such as measurement errors or natural variations, and cannot be easily reduced.

Uncertainty Quantification

Michael Panchenko

Monte Carlo-Dropout for Uncertainty Quantification in Deep Learning - Maternus Herold (appliedAI Institute)

Oct 20, 2022

Deep learning models are commonly used for regression tasks, producing a point estimate for a given set of inputs. However, a single point does not adequately reflect the model’s uncertainty about this estimate. In safety-critical applications, understanding this uncertainty can be extremely valuable. For example, in medical applications, a high degree of uncertainty could necessitate an additional check by a doctor, whereas this might not be required in cases of highly certain predictions.

Uncertainty Quantification

Maternus Herold

Sparse Random Feature Expansions for Surrogate Modelling - Maternus Herold (appliedAI Institute)

Jul 7, 2022

Surrogate models are indispensable tools for conducting uncertainty quantification and robustness analyses, thereby driving pivotal design decisions, such as those involved in automotive engineering. The necessity for surrogate models arises from the reality that simulations of an object encapsulate inherent uncertainties concerning parameters and model definitions, and these simulations are computationally costly.

Maternus Herold

Data (e)valuation and model interpretation: a game theoretic approach - Miguel de Benito Delgado

May 20, 2021

Attributing a “fair” (for some definition of fair) value to training samples has multiple applications. It can be used to investigate or improve data sources, but it can also help detect outliers and to investigate how certain features and the values thereof influence global model performance. It is also possible to improve model performance: removing points of low value can decrease model error.

Miguel de Benito Delgado

1 ...

7

... 18

Our Block Seminars

Methods and issues in Explainable AI

This seminar series explores the landscape of Explainable Artificial Intelligence (XAI) beyond the content in our training on XAI, and through a comprehensive selection of papers in the field. By the end of this series, participants will …

May 17, 2023

Uncertainty quantification for neural networks

In this seminar series, we review seminal and recent papers on uncertainty quantification that didn’t make it into our training on Bayesian ML.

Oct 22, 2022