Konrad Szafer

kszafer [at] andrew [dot] cmu [dot] edu

About

I am a research assistant at AutonLab, Carnegie Mellon University, where I am currently working on projects related to time-series foundational models and model evaluation with limited labeled data. I received my BSc in Automatic Control and Robotics with an Honors Track on Machine Learning from Poznan University of Technology in 2023. In my free time, I enjoy building open source projects and contributing to the community, playing with hardware, as well as working out outdoors and traveling.

Research

My research interests are focused on leveraging deep learning to solve complex problems across various modalities, including time-series, computer vision, and natural language processing. I place a particular emphasis on improving models’ zero-shot and transfer learning capabilities. My aim is to enhance the robustness of neural networks under uncertain conditions and to increase their interpretability. Additionally, I study the evaluation of machine learning models to ensure their safety and reliability in real-world applications.

selected publications

MOMENT: A Family of Open Time-series Foundation Models

Mononito Goswami , Konrad Szafer , Arjun Choudhry , and 3 more authors

2024

Abs HTML PDF

We introduce MOMENT, a family of open-source foundation models for general-purpose time-series analysis. Pre-training large models on time-series data is challenging due to (1) the absence of a large and cohesive public time-series repository, and (2) diverse time-series characteristics which make multi-dataset training onerous. Additionally, (3) experimental benchmarks to evaluate these models, especially in scenarios with limited resources, time, and supervision, are still in their nascent stages. To address these challenges, we compile a large and diverse collection of public time-series, called the Time-series Pile, and systematically tackle time-series-specific challenges to unlock large-scale multi-dataset pre-training. Finally, we build on recent work to design a benchmark to evaluate time-series foundation models on diverse tasks and datasets in limited supervision settings. Experiments on this benchmark demonstrate the effectiveness of our pre-trained models with minimal data and task-specific fine-tuning. Finally, we present several interesting empirical observations about large pre-trained time-series models. Our code is available anonymously at anonymous.4open.science/r/BETT-773F/.