In recent years, large language models (LLMs) have achieved remarkable performances in various NLP tasks. They can generate texts that are indistinguishable from those written by humans. Such remarkable performance of LLMs increases their risk of being used for malicious purposes, such as generating fake news articles. Therefore, it is necessary to develop methods for distinguishing texts written by LLMs from those written by humans. Watermarking is one of the most powerful methods for achieving this. Although existing watermarking methods have successfully detected texts generated by LLMs, they significantly degrade the quality of the generated texts. In this study, we propose the Necessary and Sufficient Watermark (NS-Watermark) for inserting watermarks into generated texts without degrading the text quality. More specifically, we derive minimum constraints required to be imposed on the generated texts to distinguish whether LLMs or humans write the texts. Then, we formulate the NS-Watermark as a constrained optimization problem and propose an efficient algorithm to solve it. Through the experiments, we demonstrate that the NS-Watermark can generate more natural texts than existing watermarking methods and distinguish more accurately between texts written by LLMs and those written by humans. Especially in machine translation tasks, the NS-Watermark can outperform the existing watermarking method by up to 30 BLEU scores. Large language models (LLMs) have achieved remarkable performances in a wide range of NLP tasks, including language generation (Chen et al., 2021), question answering (Joshi et al., 2017; Kwiatkowski et al., 2019), and reasoning tasks (Bisk et al., 2020; Kojima et al., 2022). Recently, many pre-trained LLMs have been released (Brown et al., 2020; Chung et al., 2022; Zhang et al., 2022; Touvron et al., 2023), which can now generate natural and fluent texts that are indistinguishable from texts written by humans.

The advent of Large Language Models (LLMs) has paved the way for complex tasks such as role-playing, which enhances user interactions by enabling models to imitate various characters. However, the closed-source nature of state-of-the-art LLMs and their general-purpose training limit role-playing optimization. In this paper, we introduce RoleLLM, a framework to benchmark, elicit, and enhance role-playing abilities in LLMs. RoleLLM comprises four stages: (1) Role Profile Construction for 100 roles; (2) Context-Based Instruction Generation (Context-Instruct) for role-specific knowledge extraction; (3) Role Prompting using GPT (RoleGPT) for speaking style imitation; and (4) Role-Conditioned Instruction Tuning (RoCIT) for fine-tuning open-source models along with role customization. By Context-Instruct and RoleGPT, we create RoleBench, the first systematic and fine-grained character-level benchmark dataset for role-playing with 168,093 samples. Moreover, RoCIT on RoleBench yields RoleLLaMA (English) and RoleGLM (Chinese), significantly enhancing role-playing abilities and even achieving comparable results with RoleGPT (using GPT-4).

Benefiting from the rapid development of deep learning, 2D and 3D computer vision applications are deployed in many safe-critical systems, such as autopilot and identity authentication. However, deep learning models are not trustworthy enough because of their limited robustness against adversarial attacks. The physically realizable adversarial attacks further pose fatal threats to the application and human safety. Lots of papers have emerged to investigate the robustness and safety of deep learning models against adversarial attacks. To lead to trustworthy AI, we first construct a general threat model from different perspectives and then comprehensively review the latest progress of both 2D and 3D adversarial attacks. We extend the concept of adversarial examples beyond imperceptive perturbations and collate over 170 papers to give an overview of deep learning model robustness against various adversarial attacks. To the best of our knowledge, we are the first to systematically investigate adversarial attacks for 3D models, a flourishing field applied to many real-world applications. In addition, we examine physical adversarial attacks that lead to safety violations. Last but not least, we summarize present popular topics, give insights on challenges, and shed light on future research on trustworthy AI.

Large language models (LLMs) with billions of parameters have demonstrated outstanding performance on various natural language processing tasks. This report presents OpenBA, an open-sourced 15B bilingual asymmetric seq2seq model, to contribute an LLM variant to the Chinese-oriented open-source model community. We enhance OpenBA with effective and efficient techniques as well as adopt a three-stage training strategy to train the model from scratch. Our solution can also achieve very competitive performance with only 380B tokens, which is better than LLaMA-70B on the BELEBELE benchmark, BLOOM-176B on the MMLU benchmark, GLM-130B on the C-Eval (hard) benchmark. This report provides the main details to pre-train an analogous model, including pre-training data processing, Bilingual Flan data collection, the empirical observations that inspire our model architecture design, training objectives of different stages, and other enhancement techniques. Additionally, we also provide the fine-tuning details of OpenBA on four downstream tasks. We have refactored our code to follow the design principles of the Huggingface Transformers Library, making it more convenient for developers to use, and released checkpoints of different training stages at https://huggingface.co/openBA.

Deep neural networks are over-parameterized and easily overfit the datasets they train on. In the extreme case, it has been shown that these networks can memorize a training set with fully randomized labels. We propose using the curvature of loss function around each training sample, averaged over training epochs, as a measure of memorization of the sample. We use this metric to study the generalization versus memorization properties of different samples in popular image datasets and show that it captures memorization statistics well, both qualitatively and quantitatively. We first show that the high curvature samples visually correspond to long-tailed, mislabeled, or conflicting samples, those that are most likely to be memorized. This analysis helps us find, to the best of our knowledge, a novel failure mode on the CIFAR100 and ImageNet datasets: that of duplicated images with differing labels. Quantitatively, we corroborate the validity of our scores via two methods. First, we validate our scores against an independent and comprehensively calculated baseline, by showing high cosine similarity with the memorization scores released by Feldman & Zhang (2020). Second, we inject corrupted samples which are memorized by the network, and show that these are learned with high curvature. An added advantage of our method is that it is scalable, as it requires training only a single network as opposed to the thousands trained by the baseline, while capturing the aforementioned failure mode that the baseline fails to identify. Deep learning has been hugely successful in many fields.

In city, information about urban objects such as water supply, railway lines, power lines, buildings, roads, etc., is necessary for city planning. In particular, information about the spread of these objects, locations and capacity is needed for the policymakers to make impactful decisions. This thesis aims to segment the building and roads from the aerial image captured by the satellites and UAVs. Many different architectures have been proposed for the semantic segmentation task and UNet being one of them. In this thesis, we propose a novel architecture based on Google's newly proposed EfficientNetV2 as an encoder for feature extraction with UNet decoder for constructing the segmentation map. Using this approach we achieved a benchmark score for the Massachusetts Building and Road dataset with an mIOU of 0.8365 and 0.9153 respectively.

Large language models (large LMs) are susceptible to producing text that contains hallucinated content. An important instance of this problem is self-contradiction, where the LM generates two contradictory sentences within the same context. In this work, we present a comprehensive investigation into self-contradiction for various instruction-tuned LMs, covering evaluation, detection, and mitigation. Our analysis reveals the prevalence of self-contradictions when LMs generate text for open-domain topics, e.g., in 17.7% of all sentences produced by ChatGPT. Self-contradiction also complements retrieval-based methods, as a large portion of them (e.g., 35.8% for ChatGPT) cannot be verified using Wikipedia. We then propose a novel prompting-based framework designed to effectively detect and mitigate self-contradictions. Our detector achieves high accuracy, e.g., around 80% F1 score when prompting ChatGPT. The mitigation algorithm iteratively refines the generated text to remove contradictory information while preserving text fluency and informativeness. Importantly, our entire framework is applicable to black-box LMs and does not require external grounded knowledge. Our approach is practically effective and has been released as a push-button tool to benefit the public, available at https://chatprotect.ai/.

Liberalism-oriented political philosophy reasons that all individuals should be treated equally independently of their protected characteristics. Related work in machine learning has translated the concept of \emph{equal treatment} into terms of \emph{equal outcome} and measured it as \emph{demographic parity} (also called \emph{statistical parity}). Our analysis reveals that the two concepts of equal outcome and equal treatment diverge; therefore, demographic parity does not faithfully represent the notion of \emph{equal treatment}. We propose a new formalization for equal treatment by (i) considering the influence of feature values on predictions, such as computed by Shapley values decomposing predictions across its features, (ii) defining distributions of explanations, and (iii) comparing explanation distributions between populations with different protected characteristics. We show the theoretical properties of our notion of equal treatment and devise a classifier two-sample test based on the AUC of an equal treatment inspector. We study our formalization of equal treatment on synthetic and natural data. We release \texttt{explanationspace}, an open-source Python package with methods and tutorials.

Language models (LMs) are becoming the foundation for almost all major language technologies, but their capabilities, limitations, and risks are not well understood. We present Holistic Evaluation of Language Models (HELM) to improve the transparency of language models. First, we taxonomize the vast space of potential scenarios (i.e. use cases) and metrics (i.e. desiderata) that are of interest for LMs. Then we select a broad subset based on coverage and feasibility, noting what's missing or underrepresented (e.g. question answering for neglected English dialects, metrics for trustworthiness). Second, we adopt a multi-metric approach: We measure 7 metrics (accuracy, calibration, robustness, fairness, bias, toxicity, and efficiency) for each of 16 core scenarios when possible (87.5% of the time). This ensures metrics beyond accuracy don't fall to the wayside, and that trade-offs are clearly exposed. We also perform 7 targeted evaluations, based on 26 targeted scenarios, to analyze specific aspects (e.g. reasoning, disinformation). Third, we conduct a large-scale evaluation of 30 prominent language models (spanning open, limited-access, and closed models) on all 42 scenarios, 21 of which were not previously used in mainstream LM evaluation. Prior to HELM, models on average were evaluated on just 17.9% of the core HELM scenarios, with some prominent models not sharing a single scenario in common. We improve this to 96.0%: now all 30 models have been densely benchmarked on the same core scenarios and metrics under standardized conditions. Our evaluation surfaces 25 top-level findings. For full transparency, we release all raw model prompts and completions publicly for further analysis, as well as a general modular toolkit. We intend for HELM to be a living benchmark for the community, continuously updated with new scenarios, metrics, and models.

This paper studies the problem of conducting self-supervised learning for node representation learning on graphs. Most existing self-supervised learning methods assume the graph is homophilous, where linked nodes often belong to the same class or have similar features. However, such assumptions of homophily do not always hold in real-world graphs. We address this problem by developing a decoupled self-supervised learning (DSSL) framework for graph neural networks. DSSL imitates a generative process of nodes and links from latent variable modeling of the semantic structure, which decouples different underlying semantics between different neighborhoods into the self-supervised learning process. Our DSSL framework is agnostic to the encoders and does not need prefabricated augmentations, thus is flexible to different graphs. To effectively optimize the framework, we derive the evidence lower bound of the self-supervised objective and develop a scalable training algorithm with variational inference. We provide a theoretical analysis to justify that DSSL enjoys the better downstream performance. Extensive experiments on various types of graph benchmarks demonstrate that our proposed framework can achieve better performance compared with competitive baselines.