Our work | appliedAI — TransferLab

Our work

We prepare in-depth reviews of proven and emerging techniques, illustrate and benchmark them with best-practices playbooks and code, and communicate them with visually rich showcases and trainings. We also maintain a blog with introductory material and paper reviews, and hold a weekly seminar where we cover both the foundations and recent advances in AI.

Our areas of interest

DevOps for ML systems

Inspired by the daily work of our engineering team, we strive to identify the challenges and solutions pertaining to development and operations in the use cases relevant to our partners and customers

Safe and efficient learning

Data-driven automation requires methods ensuring the safety of systems and their users, while remaining conservative in sample complexity.

Decision making under uncertainty

Automating decisions based on data is a multifaceted problem with many pitfalls. We tackle multiple ends of the spectrum, from the detection of anomalies to calibration of models and Bayesian decision making.

Our content series

Classifier calibration

Data valuation

Bayesian ML and Probabilistic Programming

Practical anomaly detection

Causal inference and discovery

Data efficiency

Diffusion Models

Safety in learned controllers

Explore everything

Our take on ICLR 2022 - Part 1

ICLR is one of the most important and prestigious conferences in the ML community. It spans all ML topics, from embeddings to Graph neural …

A photo of Fabio

May 10, 2022

Certified error rates for neural networks

Adversarial training has been shown to improve robustness of neural networks to certain classes of data perturbations. Despite constant …

A picture of Miguel

Robustness in ML

Jan 14, 2022

Natural, Trust Region and Proximal Policy Optimization

We present an overview of the theory behind three popular and related algorithms for gradient based policy optimization: natural policy …

A picture of Michael

Aug 10, 2021

Cross-validation: what does it estimate?

In Cross-validation: what does it estimate and how well does it do it? Bates et al. focus on cross validation for error estimation and show …

A picture of Miguel

Error Estimation and Model Selection

Jun 11, 2021

Review Partner area

Data valuation techniques. Applications, advances and problems

The attribution of value to data points can help in data collection, active learning, gauging model robustness, improving model performance …

A picture of Miguel

May 15, 2021

Review Partner area

Calibration of Classifiers

We discuss the calibration of probabilistic classifiers, i.e. the question of whether the output vectors of such classifiers accurately …

A picture of Michael

Classifier Calibration

Apr 29, 2021

What we are reading:
Paper pills

Denoising Diffusion and Score Based Generative Models

May 11, 2022

This pill gives a short overview of the development of diffusion models over the last 7 years. It contains several references useful for diving into this topic.

Diffusion Models

A photo of Fabio

Progressive Distillation for Fast Sampling of Diffusion Models

May 11, 2022

The paper describes how the generation process of diffusion models can be sped up by using a slow teacher model for training a faster student - the latter performing fewer sampling steps. This distillation process can be iterated by using the previous student as teacher and the final student was shown to generate high-quality samples in the fraction of the original model’s time.

Diffusion Models

A photo of Fabio

Comparing Distributions by Measuring Differences that Affect Decision Making

May 6, 2022

A new family of divergences for distributions is introduced, generalizing KL and others, together with an efficient estimator for it. A beautiful aspect of this family is that the divergence is based on a loss and thus can not only reflect how much distributions differ but also how much their difference affects a selected decision making scenario.

A picture of Michael

Score-Based Generative Modeling with Critically-Damped Langevin Diffusion

May 6, 2022

In this rather technical paper, a new dynamics is proposed for the sampling of diffusion models. Instead of sampling the de-noised image directly, an auxiliary velocity variable is sampled and deterministic dynamics are used to reconstruct the de-noised data. The approach outperforms previous state of the art in terms of sample quality.

Diffusion Models

A photo of Fabio

Neural Collapse in deep classifiers during Terminal Phase of Training

May 5, 2022

Neural collapse refers to the observation that the last two layers of neural networks that were trained for a long time take a very simple, restricted and universal form - namely an equiangular tight frame. Papers referenced in this pill highlight the phenomenon and provide theoretical hints for why it happens.

Robustness in ML

A picture of Michael

Come to our seminar

Data (e)valuation and model interpretation: a game theoretic approach

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.

A picture of Miguel

Introduction to simulation-based inference

Dec 17, 2020

Scientists and engineers employ stochastic numerical simulators to model empirically observed phenomena. In contrast to purely statistical models, simulators express scientific principles that provide powerful inductive biases, improve generalization to new data or scenarios and allow for fewer, more interpretable and domain-relevant parameters. Despite these advantages, tuning a simulator’s parameters so that its outputs match data is challenging.

Álvaro Tejero-Cantero

Hamiltonian Monte Carlo Sampling

Dec 3, 2020

Metropolis, Slice or Gibbs sampling tend to be inefficient when applied in high dimensions due to their local random walk behaviour. Hamiltonian Monte Carlo (HMC) was developed around a geometrical understanding of the target distribution, borrowing concepts from physics to generate transitions and exploring the target distribution. Ultimately, moving more rapidly through the target distribution.

Bayesian ML and Probabilistic Programming

Maternus Herold

Bayesian optimal experiment design (2 of 2)

Nov 26, 2020

This is the second of a two talks providing an introduction to Bayesian optimal experimental design (BOED) and how it relates to exploration in reinforcement learning and control. The first part dealt with the theoretical foundations for BOED. This talk completes the topic by presenting an application in control, which was created during my master thesis at TUDa.

Bayesian optimal experiment design (1 of 2)

Nov 19, 2020

These two talks give an introduction to Bayesian optimal experimental design (BOED) and how it relates to exploration in reinforcement learning and control. The first part deals with the theoretical foundations for BOED. Different design types are covered, e.g. space filling designs, alphabetic designs and Bayesian designs. Afterwards it is shown how Bayes' theorem relates to the mutual information (MI) or equally the expected information gain (EIG).