When working in AI, especially when doing an empirical PhD, the engineering required for effective experimentation is incredibly important. I procrastinated on writing this article for a while now, but it is getting more and more apparent these days when a large portion of research shifted to working with large foundation models. But even ~7 years ago, when I started my PhD, I quickly realized that running experiments effectively will be crucial. This is because effective experimentation means that research hypotheses can be tested quickly and provide insights into the next hypotheses to test ..read more

Are code and data available? Yes, code and data are on GitHub. Code includes both Monte Carlo conformal prediction as well as the plausibility regions from v1 of the paper. Can you derive the conformal $p$-values used in the paper? The connection of conformal prediction and $p$-values is scattered across the literature and there is, to the best of my knowledge, no good reference to understand this. So we added a thorough derivation in Appendix B of the paper. How do you get the plausibilities $\lambda$ in practice from different formats of annotations? In a nutshell, this is a modeling choice ..read more

Introduction In the beginning of my PhD, I had a series of conversations with PhD students and academics about how to keep track of everything during a PhD. Some of these discussions were motivated by the 2018 Workshop of being a Good Citizen of CVPR. This inspired me to start thinking properly about how to organize and eventually document my PhD work. In retrospective, I think this was one of my better decisions. Only recently, at the Heidelberg Laureate Forum 2023, I found that many successful academics also document, for example, who they talk to during conferences to stay on top of things ..read more

A short disclaimer is necessary before diving in: the below is a rather personal opinion on the subject — driven by my personal experiences in AI research. As such, it is not meant to blame, contradict or discredit anyone or anything. Instead it is an attempt to add color. I think the project track in question is rather specific; I am sure much thought has gone into it and NeurIPS will iterate on it in future instances of the conferences. This being said, I think many arguments raised on X are not necessarily about NeurIPS' decision to have such a track in specific. Instead, many arguments can ..read more

Introduction By construction, a PhD has a clear end. Depending on the program, country and field, a PhD is supposed to be done within 3-6 years when it is usually awarded after an official defense of the research work. This is in contrast to most other careers and jobs, especially in industry but also in the public sector. Even though a PhD is often considered as a qualification for independent research and thereby acts as the entry to an academic career, it is commonly assumed that most PhD graduates do not continue in academia. This also matches my impression and surveys among PhD students i ..read more

Ben Recht recently published some blog articles questioning the utility of prediction intervals and sets, especially as obtained using distribution-free, conformal methods. In this article, I want to add some color to the discussion given my experience with applying these methods in various settings. Let me start with the elephant in the room: do people actually want uncertainty estimates in general? If you ask people in academia or industry, the first answer is usually yes. Somehow, as researchers and engineers, we want to understand when the models we train "know" and when they do not know ..read more

Abstract Conformal Prediction (CP) allows to perform rigorous uncertainty quantification by constructing a prediction set $C(X)$ satisfying $\mathbb{P}_{agg}(Y \in C(X))\geq 1-\alpha$ for a user-chosen $\alpha \in [0,1]$ by relying on calibration data $(X_1,Y_1),...,(X_n,Y_n)$ from $\mathbb{P}=\mathbb{P}_{agg}^{X} \otimes \mathbb{P}_{agg}^{Y|X}$. It is typically implicitly assumed that $\mathbb{P}_{agg}^{Y|X}$ is the ``true'' posterior label distribution. However, in many real-world scenarios, the labels $Y_1,...,Y_n$ are obtained by aggregating expert opinions using a voting procedure, result ..read more

Abstract For safety, AI systems in health undergo thorough evaluations before deployment, validating their predictions against a ground truth that is assumed certain. However, this is actually not the case and the ground truth may be uncertain. Unfortunately, this is largely ignored in standard evaluation of AI models but can have severe consequences such as overestimating the future performance. To avoid this, we measure the effects of ground truth uncertainty, which we assume decomposes into two main components: annotation uncertainty which stems from the lack of reliable annotations, and in ..read more

Abstract For safety, AI systems in health undergo thorough evaluations before deployment, validating their predictions against a ground truth that is assumed certain. However, this is actually not the case and the ground truth may be uncertain. Unfortunately, this is largely ignored in standard evaluation of AI models but can have severe consequences such as overestimating the future performance. To avoid this, we measure the effects of ground truth uncertainty, which we assume decomposes into two main components: annotation uncertainty which stems from the lack of reliable annotations, and in ..read more

Letitia is a PhD student at Heidelberg University in the Natural Language Processing group working on vision-language models. On top of her research, she is running a YouTube channel covering Ai papers and developments. So it was a pleasure to be interviewed for her channel about my PhD research on adversarial robustness: The post Interviewed by AI Coffee Break with Letitia appeared first on David Stutz ..read more