A new arXiv paper examines pretraining methods for neural decoders used in brain-computer interfaces. Because training a high-performing decoder normally requires large labeled datasets from each new subject, the work targets ways to lower that annotation burden. The abstract is truncated in the source, so full results are not yet available.
A new arXiv preprint in machine learning asks whether manual feature engineering remains necessary now that tabular foundation models exist. These models are pretrained across many tabular datasets and applied through in-context learning, which may reduce reliance on hand-crafted features. The abstract frames the question as an open issue for tabular machine learning research.
A new arXiv preprint explores whether midtraining, a stage between pretraining and post-training, can influence which behaviors a large language model carries forward. The authors propose a method called Inoculation Midtraining, which uses invented words to shape how desirable and undesirable properties generalize. The work is a research contribution and has not been peer reviewed.
A new arXiv preprint looks at how time series foundation models are pretrained on artificially generated trajectories where the underlying data-generating process is known. The work focuses on distillation methods rather than the conventional loss-based pretraining objectives that compare model outputs against targets. It aims to improve how these models learn from synthetic time series data.