SCHEDULE
Propagating Surrogate Uncertainty in Bayesian Inverse Problems
by Andrew Roberts
Standard Bayesian inference schemes are infeasible for inverse problems with computationally expensive forward models. A common solution is to replace the model with a cheaper surrogate. To avoid overconfident conclusions, it is essential to acknowledge the surrogate approximation by propagating its uncertainty. At present, a variety of distinct uncertainty propagation methods have been suggested, with little understanding of how they vary. To fill this gap, we propose a mixture distribution termed the expected posterior (EP) as a general baseline for uncertainty-aware posterior approximation, justified by decision theoretic and modular Bayesian inference arguments. We compare this distribution to popular alternatives, present an approximate Markov chain Monte Carlo sampler for EP-based inference, and highlight future directions.
TBD
by Tejovan Parker
Details will be updated soon.
Western Pacific tropical cyclones over the past 500 years: when a deep-learning climate emulator meets a Chinese handwritten historical record
by Mu-Ting Chien
Digitized handwritten Chinese historical records REACHES show that tropical cyclone (TC) landfall frequency peaked in 1650-1680 AD over the past 500 years. However, the environmental conditions that lead to this peak remain unknown. This study uses a novel deep-learning climate emulator, ACE2, and a dynamical model, HiRAM, both forced with the last-millennium reconstructed sea surface temperatures and sea ice to uncover the large-scale climate states that drive the long-term variability in Western Pacific TC frequency and track. We find that simulated TC landfall frequency in East Asia also peaks in ACE2 during the 1650-1680 AD period, consistent with REACHES data. Furthermore, the seasonal cycle of Western Pacific TC activity has two peaks during this period, different from a single peak in the current climate, possibly associated with the shift from the East Asian monsoon to the South Asian monsoon. We investigate the large-scale circulation and environmental conditions that drive changes in TC genesis, track, and seasonal cycle over the past 500 years. Our lessons learned have implications for future changes in TC activities in the Western Pacific. Meanwhile, our work proposes a framework to investigate paleoclimate TCs by combining an AI global climate emulator with proxy data.
TBD
by Freddy Reiber & Tyler Calabrese
Details will be updated soon.
TBD
by Micah Benson & Clark Ikezu
Details will be updated soon.
Stop the Nonconsensual Use of Nude Images in Research (Published at NeurIPS 2025 - Oral)
by Princessa Cintaqia
In order to train, test, and evaluate nudity detection models, machine learning researchers typically rely on nude images scraped from the Internet. Our research finds that this content is collected and, in some cases, subsequently distributed by researchers without consent, leading to potential misuse and exacerbating harm against the subjects depicted. We argue that the distribution of nonconsensually collected nude images by researchers perpetuates image-based sexual abuse and that the machine learning community should stop the nonconsensual use of nude images in research. To characterize the scope and nature of this problem, we conducted a systematic review of papers published in computing venues that collect and use nude images. Our results paint a grim reality: norms around the usage of nude images are sparse, leading to a litany of problematic practices like distributing and publishing nude images with uncensored faces, and intentionally collecting and sharing abusive content. We conclude with a call-to-action for publishing venues and a vision for research in nudity detection that balances user agency with concrete research objectives. You can check out the paper here: openreview.net/pdf?id=Ev5xwr3vWh