Search

Pretrained large language models (LLMs) have revolutionized natural language processing (NLP) tasks such as summarization, question answering, and translation. However, LLMs pose significant security risks due to their tendency to memorize training data, leading to potential privacy breaches and copyright infringement. Accurate measurement of this memorization is essential to evaluate and mitigate these potential risks. However, previous attempts to characterize memorization are constrained by either using prefixes only or by prepending a constant soft prompt to the prefixes, which cannot react to changes in input. To address this challenge, we propose a novel method for estimating LLM memorization using dynamic, prefix-dependent soft prompts. Our approach involves training a transformer-based generator to produce soft prompts that adapt to changes in input, thereby enabling more accurate extraction of memorized data. Our method not only addresses the limitations of previous methods but also demonstrates superior performance in diverse experimental settings compared to state-of-the-art techniques. In particular, our method can achieve the maximum relative improvement of 112.75% and 32.26% over the vanilla baseline in terms of discoverable memorization rate for the text generation task and code generation task respectively.

Pulsar timing is a powerful tool that, by accounting for every rotation of a pulsar, precisely measures the spin frequency, spin frequency derivatives, astrometric position, binary parameters when applicable, properties of the ISM, and potentially general relativistic effects. Typically, this process demands fairly stringent scheduling requirements for monitoring observations as well as deep domain knowledge to "phase connect" the timing data. We present an algorithm that automates the pulsar timing process for binary pulsars, whose timing solutions have an additional level of complexity, although the algorithm works for isolated pulsars as well. Using the statistical F-test and the quadratic dependence of the reduced $\chi ^2$ near a minimum, the global rotation count of a pulsar can be determined efficiently and systematically. We have used our algorithm to establish timing solutions for two newly discovered binary pulsars, PSRs J1748$-$2446aq and J1748$-$2446at, in the globular cluster Terzan 5, using $\sim$70 Green Bank Telescope observations from the last 13 years., Comment: 12 pages, 5 figures, 1 table; published in ApJ

Students of visualization come to formal education with an abundance of personal experience. However, one's exposure to graphics through media and education may not be sufficiently diverse to appreciate the nuance and complexity required to design and evaluate effective representations. While many introductory courses in visualization address best practices for visual encoding of data based on perceptual characteristics, as cognitive scientists, we place equal value on representational decisions based on communicative context: how the representation is intended to be used. In this pedagogical activity, we aim to surface learners' preconceived notions about what makes a visualization effective. Here we describe the structure and context of an introductory-level visualization activity, how it might be conducted in individual or group settings, our experience with the common misconceptions the activity can reveal, and conclude with recommendations on how they might be addressed.

Offers seven lesson plans to help blend the affective and cognitive realms of children's learning, connecting an emotional lexicon with written and spoken dialogs, photographs, arguments, and movements. (DMM)