Hallucination in Natural Language Generation (NLG) is like the elephant in the room, obvious but often overlooked until recent achievements significantly improved the fluency and grammatical accuracy of generated text. For Large Language Models (LLMs), hallucinations can happen in various downstream tasks and casual conversations, which need accurate assessment to enhance reliability and safety. However, current studies on hallucination evaluation vary greatly, and people still find it difficult to sort out and select the most appropriate evaluation methods. Moreover, as NLP research gradually shifts to the domain of LLMs, it brings new challenges to this direction. This paper provides a comprehensive survey on the evolvement of hallucination evaluation methods, aiming to address three key aspects: 1) Diverse definitions and granularity of facts; 2) The categories of automatic evaluators and their applicability; 3) Unresolved issues and future directions.
相關內容
Research on Large Language Models (LLMs) has often neglected subtle biases that, although less apparent, can significantly influence the models' outputs toward particular social narratives. This study addresses two such biases within LLMs: representative bias, which denotes a tendency of LLMs to generate outputs that mirror the experiences of certain identity groups, and affinity bias, reflecting the models' evaluative preferences for specific narratives or viewpoints. We introduce two novel metrics to measure these biases: the Representative Bias Score (RBS) and the Affinity Bias Score (ABS), and present the Creativity-Oriented Generation Suite (CoGS), a collection of open-ended tasks such as short story writing and poetry composition, designed with customized rubrics to detect these subtle biases. Our analysis uncovers marked representative biases in prominent LLMs, with a preference for identities associated with being white, straight, and men. Furthermore, our investigation of affinity bias reveals distinctive evaluative patterns within each model, akin to `bias fingerprints'. This trend is also seen in human evaluators, highlighting a complex interplay between human and machine bias perceptions.
Large Language Models (LLMs) have demonstrated remarkable success in tasks like the Winograd Schema Challenge (WSC), showcasing advanced textual common-sense reasoning. However, applying this reasoning to multimodal domains, where understanding text and images together is essential, remains a substantial challenge. To address this, we introduce WinoVis, a novel dataset specifically designed to probe text-to-image models on pronoun disambiguation within multimodal contexts. Utilizing GPT-4 for prompt generation and Diffusion Attentive Attribution Maps (DAAM) for heatmap analysis, we propose a novel evaluation framework that isolates the models' ability in pronoun disambiguation from other visual processing challenges. Evaluation of successive model versions reveals that, despite incremental advancements, Stable Diffusion 2.0 achieves a precision of 56.7% on WinoVis, only marginally surpassing random guessing. Further error analysis identifies important areas for future research aimed at advancing text-to-image models in their ability to interpret and interact with the complex visual world.
As the use of Large Language Models (LLMs) becomes more widespread, understanding their self-evaluation of confidence in generated responses becomes increasingly important as it is integral to the reliability of the output of these models. We introduce the concept of Confidence-Probability Alignment, that connects an LLM's internal confidence, quantified by token probabilities, to the confidence conveyed in the model's response when explicitly asked about its certainty. Using various datasets and prompting techniques that encourage model introspection, we probe the alignment between models' internal and expressed confidence. These techniques encompass using structured evaluation scales to rate confidence, including answer options when prompting, and eliciting the model's confidence level for outputs it does not recognize as its own. Notably, among the models analyzed, OpenAI's GPT-4 showed the strongest confidence-probability alignment, with an average Spearman's $\hat{\rho}$ of 0.42, across a wide range of tasks. Our work contributes to the ongoing efforts to facilitate risk assessment in the application of LLMs and to further our understanding of model trustworthiness.
The burgeoning interest in developing Large Language Models (LLMs) with up to trillion parameters has been met with concerns regarding resource efficiency and practical expense, particularly given the immense cost of experimentation. This scenario underscores the importance of exploring the potential of Small Language Models (SLMs) as a resource-efficient alternative. In this context, we introduce MiniCPM, specifically the 1.2B and 2.4B non-embedding parameter variants, not only excel in their respective categories but also demonstrate capabilities on par with 7B-13B LLMs. While focusing on SLMs, our approach exhibits scalability in both model and data dimensions for future LLM research. Regarding model scaling, we employ extensive model wind tunnel experiments for stable and optimal scaling. For data scaling, we introduce a Warmup-Stable-Decay (WSD) learning rate scheduler (LRS), conducive to continuous training and domain adaptation. We present an in-depth analysis of the intriguing training dynamics that occurred in the WSD LRS. With WSD LRS, we are now able to efficiently study data-model scaling law without extensive retraining experiments on both axes of model and data, from which we derive the much higher compute optimal data-model ratio than Chinchilla Optimal. Additionally, we introduce MiniCPM family, including MiniCPM-DPO, MiniCPM-MoE and MiniCPM-128K, whose excellent performance further cementing MiniCPM's foundation in diverse SLM applications. MiniCPM models are available publicly at //github.com/OpenBMB/MiniCPM .
Residential fixed broadband internet access in the United States (US) has long been distributed inequitably, drawing significant attention from researchers and policymakers. This paper evaluates the efficacy of the Connect America Fund (CAF), a key policy intervention aimed at addressing disparities in US internet access. CAF subsidizes the creation of new regulated broadband monopolies in underserved areas, aiming to provide comparable internet access, in terms of price and speed, to that available in urban regions. Oversight of CAF largely relies on data self-reported by internet service providers (ISPs), which is often questionable. We use the broadband-plan querying tool (BQT) to curate a novel dataset that complements ISP-reported information with ISP-advertised broadband plan details (download speed and monthly cost) on publicly accessible websites. Specifically, we query advertised broadband plans for 687k residential addresses across 15 states, certified as served by ISPs to regulators. Our analysis reveals significant discrepancies between ISP-reported data and actual broadband availability. We find that the serviceability rate-defined as the fraction of addresses ISPs actively serve out of the total queried, weighted by the number of CAF addresses in a census block group-is only 55%, dropping to as low as 18% in some states. Additionally, the compliance rate-defined as the weighted fraction of addresses where ISPs actively serve and advertise download speeds above the FCC's 10 Mbps threshold-is only 33%. We also observe that in a subset of census blocks, CAF-funded addresses receive higher broadband speeds than their monopoly-served neighbors. These results indicate that while a few users have benefited from this multi-billion dollar program, it has largely failed to achieve its intended goal, leaving many targeted rural communities with inadequate or no broadband connectivity.
Scaled Dot Product Attention (SDPA) is the backbone of many modern deep-learning models. It is so versatile that it has been used in natural language, vision, and multi-modal domains with very little change compared to its original formulation. This paper discusses why the current formulation is inefficient by delving into the mathematical details of the attention mechanism. We propose three improvements to mitigate these inefficiencies, thereby, introducing three enhanced attention mechanisms: Optimised, Efficient, and Super Attention. Optimised and Efficient Attention have one and two matrix multiplications fewer per head, respectively, and 25% and 50% fewer parameters, respectively, than standard SDPA, but perform similarly to standard SDPA in both vision and natural language tasks. They can be used in all applications where SDPA is used while offering smaller model sizes and faster training and inference without noticeable loss in performance. Super Attention introduces a new linear transformation on the values, transforming them from the left. It outperforms standard SPDA on vision and natural language tasks by up to 17% while having one fewer matrix multiplication per head and 25% fewer parameters than standard SDPA. Consequently, it is also faster than standard SDPA. Super Attention is ideal in applications where the attention layer's context length is fixed, such as Vision Transformers. In addition to providing mathematical reasoning, we evaluate the presented attention mechanisms on several datasets including MNIST, CIFAR100, ImageNet, IMDB Movie Reviews, and Amazon Reviews datasets, as well as combined Europarl and Anki English-Spanish datasets for neural machine translation.
As Large Language Models (LLMs) have made significant advancements across various tasks, such as question answering, translation, text summarization, and dialogue systems, the need for accuracy in information becomes crucial, especially for serious financial products serving billions of users like Alipay. However, for a real-world product serving millions of users, the inference speed of LLMs becomes a critical factor compared to a mere experimental model. Hence, this paper presents a generic framework for accelerating the inference process, resulting in a substantial increase in speed and cost reduction for our LLM-based scenarios, with lossless generation accuracy. In the traditional inference process, each token is generated sequentially by the LLM, leading to a time consumption proportional to the number of generated tokens. To enhance this process, our framework, named \textit{lookahead}, introduces a \textit{multi-branch} strategy. Instead of generating a single token at a time, we propose a Trie-based retrieval and verification mechanism to be able to accept several tokens at a forward step. Our strategy offers two distinct advantages: (1) it guarantees absolute correctness of the output, avoiding any approximation algorithms, and (2) the worst-case performance of our approach is equivalent to the conventional process. We conduct extensive experiments to demonstrate the significant improvements achieved by applying our inference acceleration framework. Our framework is widely deployed in Alipay since April 2023, and obtain remarkable 2.66x to 6.26x speedup. Our code is available at //github.com/alipay/PainlessInferenceAcceleration.
DNS Security Extensions (DNSSEC) provide the most effective way to fight DNS cache poisoning attacks. Yet, very few DNS resolvers perform DNSSEC validation. Identifying such systems is non-trivial and the existing methods are not suitable for Internet-scale measurements. In this paper, we propose a novel remote technique for identifying DNSSEC-validating resolvers. The proposed method consists of two steps. In the first step, we identify open resolvers by scanning 3.1 billion end hosts and request every non-forwarder to resolve one correct and seven deliberately misconfigured domains. We then build a classifier that discriminates validators from non-validators based on query patterns and DNS response codes. We find that while most open resolvers are DNSSEC-enabled, less than 18% in IPv4 (38% in IPv6) validate received responses. In the second step, we remotely identify closed non-forwarders in networks that do not have inbound Source Address Validation (SAV) in place. Using the classifier built in step one, we identify 37.4% IPv4 (42.9% IPv6) closed DNSSEC validators and cross-validate the results using RIPE Atlas probes. Finally, we show that the discovered (non)-validators actively send requests to DNS root servers, suggesting that we deal with operational recursive resolvers rather than misconfigured machines.
Connecting Vision and Language plays an essential role in Generative Intelligence. For this reason, in the last few years, a large research effort has been devoted to image captioning, i.e. the task of describing images with syntactically and semantically meaningful sentences. Starting from 2015 the task has generally been addressed with pipelines composed of a visual encoding step and a language model for text generation. During these years, both components have evolved considerably through the exploitation of object regions, attributes, and relationships and the introduction of multi-modal connections, fully-attentive approaches, and BERT-like early-fusion strategies. However, regardless of the impressive results obtained, research in image captioning has not reached a conclusive answer yet. This work aims at providing a comprehensive overview and categorization of image captioning approaches, from visual encoding and text generation to training strategies, used datasets, and evaluation metrics. In this respect, we quantitatively compare many relevant state-of-the-art approaches to identify the most impactful technical innovations in image captioning architectures and training strategies. Moreover, many variants of the problem and its open challenges are analyzed and discussed. The final goal of this work is to serve as a tool for understanding the existing state-of-the-art and highlighting the future directions for an area of research where Computer Vision and Natural Language Processing can find an optimal synergy.
Deep Learning algorithms have achieved the state-of-the-art performance for Image Classification and have been used even in security-critical applications, such as biometric recognition systems and self-driving cars. However, recent works have shown those algorithms, which can even surpass the human capabilities, are vulnerable to adversarial examples. In Computer Vision, adversarial examples are images containing subtle perturbations generated by malicious optimization algorithms in order to fool classifiers. As an attempt to mitigate these vulnerabilities, numerous countermeasures have been constantly proposed in literature. Nevertheless, devising an efficient defense mechanism has proven to be a difficult task, since many approaches have already shown to be ineffective to adaptive attackers. Thus, this self-containing paper aims to provide all readerships with a review of the latest research progress on Adversarial Machine Learning in Image Classification, however with a defender's perspective. Here, novel taxonomies for categorizing adversarial attacks and defenses are introduced and discussions about the existence of adversarial examples are provided. Further, in contrast to exisiting surveys, it is also given relevant guidance that should be taken into consideration by researchers when devising and evaluating defenses. Finally, based on the reviewed literature, it is discussed some promising paths for future research.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191