ChemEval: A Comprehensive Multi-Level Chemical Evaluation for Large Language Models

There is a growing interest in the role that LLMs play in chemistry which lead to an increased focus on the development of LLMs benchmarks tailored to chemical domains to assess the performance of LLMs across a spectrum of chemical tasks varying in type and complexity. However, existing benchmarks in this domain fail to adequately meet the specific requirements of chemical research professionals. To this end, we propose \textbf{\textit{ChemEval}}, which provides a comprehensive assessment of the capabilities of LLMs across a wide range of chemical domain tasks. Specifically, ChemEval identified 4 crucial progressive levels in chemistry, assessing 12 dimensions of LLMs across 42 distinct chemical tasks which are informed by open-source data and the data meticulously crafted by chemical experts, ensuring that the tasks have practical value and can effectively evaluate the capabilities of LLMs. In the experiment, we evaluate 12 mainstream LLMs on ChemEval under zero-shot and few-shot learning contexts, which included carefully selected demonstration examples and carefully designed prompts. The results show that while general LLMs like GPT-4 and Claude-3.5 excel in literature understanding and instruction following, they fall short in tasks demanding advanced chemical knowledge. Conversely, specialized LLMs exhibit enhanced chemical competencies, albeit with reduced literary comprehension. This suggests that LLMs have significant potential for enhancement when tackling sophisticated tasks in the field of chemistry. We believe our work will facilitate the exploration of their potential to drive progress in chemistry. Our benchmark and analysis will be available at {\color{blue} \url{https://github.com/USTC-StarTeam/ChemEval}}.

Soli-enabled Noncontact Heart Rate Detection for Sleep and Meditation Tracking

Heart rate (HR) is a crucial physiological signal that can be used to monitor health and fitness. Traditional methods for measuring HR require wearable devices, which can be inconvenient or uncomfortable, especially during sleep and meditation. Noncontact HR detection methods employing microwave radar can be a promising alternative. However, the existing approaches in the literature usually use high-gain antennas and require the sensor to face the user's chest or back, making them difficult to integrate into a portable device and unsuitable for sleep and meditation tracking applications. This study presents a novel approach for noncontact HR detection using a miniaturized Soli radar chip embedded in a portable device (Google Nest Hub). The chip has a $6.5 \mbox{ mm} \times 5 \mbox{ mm} \times 0.9 \mbox{ mm}$ dimension and can be easily integrated into various devices. The proposed approach utilizes advanced signal processing and machine learning techniques to extract HRs from radar signals. The approach is validated on a sleep dataset (62 users, 498 hours) and a meditation dataset (114 users, 1131 minutes). The approach achieves a mean absolute error (MAE) of $1.69$ bpm and a mean absolute percentage error (MAPE) of $2.67\%$ on the sleep dataset. On the meditation dataset, the approach achieves an MAE of $1.05$ bpm and a MAPE of $1.56\%$. The recall rates for the two datasets are $88.53\%$ and $98.16\%$, respectively. This study represents the first application of the noncontact HR detection technology to sleep and meditation tracking, offering a promising alternative to wearable devices for HR monitoring during sleep and meditation.

Towards a Personal Health Large Language Model

In health, most large language model (LLM) research has focused on clinical tasks. However, mobile and wearable devices, which are rarely integrated into such tasks, provide rich, longitudinal data for personal health monitoring. Here we present Personal Health Large Language Model (PH-LLM), fine-tuned from Gemini for understanding and reasoning over numerical time-series personal health data. We created and curated three datasets that test 1) production of personalized insights and recommendations from sleep patterns, physical activity, and physiological responses, 2) expert domain knowledge, and 3) prediction of self-reported sleep outcomes. For the first task we designed 857 case studies in collaboration with domain experts to assess real-world scenarios in sleep and fitness. Through comprehensive evaluation of domain-specific rubrics, we observed that Gemini Ultra 1.0 and PH-LLM are not statistically different from expert performance in fitness and, while experts remain superior for sleep, fine-tuning PH-LLM provided significant improvements in using relevant domain knowledge and personalizing information for sleep insights. We evaluated PH-LLM domain knowledge using multiple choice sleep medicine and fitness examinations. PH-LLM achieved 79% on sleep and 88% on fitness, exceeding average scores from a sample of human experts. Finally, we trained PH-LLM to predict self-reported sleep quality outcomes from textual and multimodal encoding representations of wearable data, and demonstrate that multimodal encoding is required to match performance of specialized discriminative models. Although further development and evaluation are necessary in the safety-critical personal health domain, these results demonstrate both the broad knowledge and capabilities of Gemini models and the benefit of contextualizing physiological data for personal health applications as done with PH-LLM.

Ignore Me But Don't Replace Me: Utilizing Non-Linguistic Elements for Pretraining on the Cybersecurity Domain

Cybersecurity information is often technically complex and relayed through unstructured text, making automation of cyber threat intelligence highly challenging. For such text domains that involve high levels of expertise, pretraining on in-domain corpora has been a popular method for language models to obtain domain expertise. However, cybersecurity texts often contain non-linguistic elements (such as URLs and hash values) that could be unsuitable with the established pretraining methodologies. Previous work in other domains have removed or filtered such text as noise, but the effectiveness of these methods have not been investigated, especially in the cybersecurity domain. We propose different pretraining methodologies and evaluate their effectiveness through downstream tasks and probing tasks. Our proposed strategy (selective MLM and jointly training NLE token classification) outperforms the commonly taken approach of replacing non-linguistic elements (NLEs). We use our domain-customized methodology to train CyBERTuned, a cybersecurity domain language model that outperforms other cybersecurity PLMs on most tasks.

GI-PIP: Do We Require Impractical Auxiliary Dataset for Gradient Inversion Attacks?

Deep gradient inversion attacks expose a serious threat to Federated Learning (FL) by accurately recovering private data from shared gradients. However, the state-of-the-art heavily relies on impractical assumptions to access excessive auxiliary data, which violates the basic data partitioning principle of FL. In this paper, a novel method, Gradient Inversion Attack using Practical Image Prior (GI-PIP), is proposed under a revised threat model. GI-PIP exploits anomaly detection models to capture the underlying distribution from fewer data, while GAN-based methods consume significant more data to synthesize images. The extracted distribution is then leveraged to regulate the attack process as Anomaly Score loss. Experimental results show that GI-PIP achieves a 16.12 dB PSNR recovery using only 3.8% data of ImageNet, while GAN-based methods necessitate over 70%. Moreover, GI-PIP exhibits superior capability on distribution generalization compared to GAN-based methods. Our approach significantly alleviates the auxiliary data requirement on both amount and distribution in gradient inversion attacks, hence posing more substantial threat to real-world FL.

Malla: Demystifying Real-world Large Language Model Integrated Malicious Services

The underground exploitation of large language models (LLMs) for malicious services (i.e., Malla) is witnessing an uptick, amplifying the cyber threat landscape and posing questions about the trustworthiness of LLM technologies. However, there has been little effort to understand this new cybercrime, in terms of its magnitude, impact, and techniques. In this paper, we conduct the first systematic study on 212 real-world Mallas, uncovering their proliferation in underground marketplaces and exposing their operational modalities. Our study discloses the Malla ecosystem, revealing its significant growth and impact on today's public LLM services. Through examining 212 Mallas, we uncovered eight backend LLMs used by Mallas, along with 182 prompts that circumvent the protective measures of public LLM APIs. We further demystify the tactics employed by Mallas, including the abuse of uncensored LLMs and the exploitation of public LLM APIs through jailbreak prompts. Our findings enable a better understanding of the real-world exploitation of LLMs by cybercriminals, offering insights into strategies to counteract this cybercrime.

DarkBERT: A Language Model for the Dark Side of the Internet

Recent research has suggested that there are clear differences in the language used in the Dark Web compared to that of the Surface Web. As studies on the Dark Web commonly require textual analysis of the domain, language models specific to the Dark Web may provide valuable insights to researchers. In this work, we introduce DarkBERT, a language model pretrained on Dark Web data. We describe the steps taken to filter and compile the text data used to train DarkBERT to combat the extreme lexical and structural diversity of the Dark Web that may be detrimental to building a proper representation of the domain. We evaluate DarkBERT and its vanilla counterpart along with other widely used language models to validate the benefits that a Dark Web domain specific model offers in various use cases. Our evaluations show that DarkBERT outperforms current language models and may serve as a valuable resource for future research on the Dark Web.

OpenContrails: Benchmarking Contrail Detection on GOES-16 ABI

Contrails (condensation trails) are line-shaped ice clouds caused by aircraft and are likely the largest contributor of aviation-induced climate change. Contrail avoidance is potentially an inexpensive way to significantly reduce the climate impact of aviation. An automated contrail detection system is an essential tool to develop and evaluate contrail avoidance systems. In this paper, we present a human-labeled dataset named OpenContrails to train and evaluate contrail detection models based on GOES-16 Advanced Baseline Imager (ABI) data. We propose and evaluate a contrail detection model that incorporates temporal context for improved detection accuracy. The human labeled dataset and the contrail detection outputs are publicly available on Google Cloud Storage at gs://goes_contrails_dataset.

Towards Best Practice of Interpreting Deep Learning Models for EEG-based Brain Computer Interfaces

Understanding deep learning models is important for EEG-based brain-computer interface (BCI), since it not only can boost trust of end users but also potentially shed light on reasons that cause a model to fail. However, deep learning interpretability has not yet raised wide attention in this field. It remains unknown how reliably existing interpretation techniques can be used and to which extent they can reflect the model decisions. In order to fill this research gap, we conduct the first quantitative evaluation and explore the best practice of interpreting deep learning models designed for EEG-based BCI. We design metrics and test seven well-known interpretation techniques on benchmark deep learning models. Results show that methods of GradientInput, DeepLIFT, integrated gradient, and layer-wise relevance propagation (LRP) have similar and better performance than saliency map, deconvolution and guided backpropagation methods for interpreting the model decisions. In addition, we propose a set of processing steps that allow the interpretation results to be visualized in an understandable and trusted way. Finally, we illustrate with samples on how deep learning interpretability can benefit the domain of EEG-based BCI. Our work presents a promising direction of introducing deep learning interpretability to EEG-based BCI.

Hetero-SCAN: Towards Social Context Aware Fake News Detection via Heterogeneous Graph Neural Network

Fake news, false or misleading information presented as news, has a great impact on many aspects of society, such as politics and healthcare. To handle this emerging problem, many fake news detection methods have been proposed, applying Natural Language Processing (NLP) techniques on the article text. Considering that even people cannot easily distinguish fake news by news content, these text-based solutions are insufficient. To further improve fake news detection, researchers suggested graph-based solutions, utilizing the social context information such as user engagement or publishers information. However, existing graph-based methods still suffer from the following four major drawbacks: 1) expensive computational cost due to a large number of user nodes in the graph, 2) the error in sub-tasks, such as textual encoding or stance detection, 3) loss of rich social context due to homogeneous representation of news graphs, and 4) the absence of temporal information utilization. In order to overcome the aforementioned issues, we propose a novel social context aware fake news detection method, Hetero-SCAN, based on a heterogeneous graph neural network. Hetero-SCAN learns the news representation from the heterogeneous graph of news in an end-to-end manner. We demonstrate that Hetero-SCAN yields significant improvement over state-of-the-art text-based and graph-based fake news detection methods in terms of performance and efficiency.