Watermarking techniques embed statistical signals within content generated by large language models to help trace its source. Although existing methods perform well on long texts, their effectiveness significantly decreases for shorter texts. We introduce a statistical detection approach that improves the power of watermark detection, particularly in shorter texts. Our method leverages both the watermark key sequence and the next token probabilities (NTPs) to determine whether a text is generated by a large language model. We demonstrate the optimality of our approach and analyze its power properties. We also investigate an approach to estimating NTPs and extend our method to scenarios where texts face potential attacks such as substitutions, insertions, or deletions. We validate the effectiveness of our technique using texts generated by Meta-Llama-3-8B from Meta and Mistral-7B-v0.1 from Mistral AI, utilizing prompts extracted from Google’s C4 dataset. In scenarios without attacks and with short text lengths, our method demonstrates approximately 65% power improvement compared to the baseline method on average. We release all code publicly at \url{https://github.com/doccstat/llm-watermark-adaptive.}

Watermarking is a technique that involves embedding nearly unnoticeable statistical signals within generated content to help trace its source. This work focuses on a scenario where an untrusted third-party user sends prompts to a trusted language model (LLM) provider, who then generates a text from their LLM with a watermark. This setup makes it possible for a detector to later identify the source of the text if the user publishes it. The user can modify the generated text by substitutions, insertions, or deletions. Our objective is to develop a statistical method to detect if a published text is LLM-generated from the perspective of a detector. We further propose a methodology to segment the published text into watermarked and non-watermarked sub-strings. The proposed approach is built upon randomization tests and change point detection techniques. We demonstrate that our method ensures Type I and Type II error control and can accurately identify watermarked sub-strings by finding the corresponding change point locations. To validate our technique, we apply it to texts generated by several language models with prompts extracted from Google's C4 dataset and obtain encouraging numerical results. We release all code publicly at https://github.com/doccstat/llm-watermark-cpd.

Cell-surface proteins play a critical role in cell function and are primary targets for therapeutics. CITE-seq is a single-cell technique that enables simultaneous measurement of gene and surface protein expression. It is powerful but costly and technically challenging. Computational methods have been developed to predict surface protein expression using gene expression information such as from single-cell RNA sequencing (scRNA-seq) data. Existing methods however are computationally demanding and lack the interpretability to reveal underlying biological processes. We propose CrossmodalNet, an interpretable machine learning model, to predict surface protein expression from scRNA-seq data. Our model with a customized adaptive loss accurately predicts surface protein abundances. When samples from multiple time points are given, our model encodes temporal information into an easy-to-interpret time embedding to make prediction in a time-point-specific manner, and is able to uncover noise-free causal gene-protein relationships. Using three publicly available time-resolved CITE-seq data sets, we validate the performance of our model by comparing it with benchmarking methods and evaluate its interpretability. Together, we show that our method accurately and interpretably profiles surface protein expression using scRNA-seq data, thereby expanding the capacity of CITE-seq experiments for investigating molecular mechanisms involving surface proteins.