As language models are adapted by a more sophisticated and diverse set of users, the importance of guaranteeing that they provide factually correct information supported by verifiable sources is critical across fields of study & professions. This is especially the case for high-stakes fields, such as medicine and law, where the risk of propagating false information is high and can lead to undesirable societal consequences. Previous work studying factuality and attribution has not focused on analyzing these characteristics of language model outputs in domain-specific scenarios. In this work, we present an evaluation study analyzing various axes of factuality and attribution provided in responses from a few systems, by bringing domain experts in the loop. Specifically, we first collect expert-curated questions from 484 participants across 32 fields of study, and then ask the same experts to evaluate generated responses to their own questions. We also ask experts to revise answers produced by language models, which leads to ExpertQA, a high-quality long-form QA dataset with 2177 questions spanning 32 fields, along with verified answers and attributions for claims in the answers.
相關內容
Sound over-approximation methods have been proved effective for guaranteeing the absence of errors, but inevitably they produce false alarms that can hamper the programmers. Conversely, under-approximation methods are aimed at bug finding and are free from false alarms. We introduce Sufficient Incorrectness Logic (SIL), a new under-approximating, triple-based program logic to reason about program errors. SIL is designed to set apart the initial states leading to errors. We prove that SIL is correct and complete for a minimal set of rules, and we study additional rules that can facilitate program analyses. We formally compare SIL to existing triple-based program logics. Incorrectness Logic and SIL both perform under-approximations, but while the former exposes only true errors, the latter locates the set of initial states that lead to such errors, as Outcome Logic can do too. Hoare Logic performs over-approximations and as such cannot capture the set of initial states leading to errors in nondeterministic programs -- for deterministic and terminating programs, Hoare Logic and SIL coincide. Finally, we instantiate SIL with Separation Logic formulae (Separation SIL) to handle pointers and dynamic allocation and we prove its correctness. We argue that in some cases Separation SIL can yield more succinct postconditions and provide stronger guarantees than Incorrectness Separation Logic and can support effective backward reasoning.
Modern language models (LMs) have been successfully employed in source code generation and understanding, leading to a significant increase in research focused on learning-based code intelligence, such as automated bug repair, and test case generation. Despite their great potential, language models for code intelligence (LM4Code) are susceptible to potential pitfalls, which hinder realistic performance and further impact their reliability and applicability in real-world deployment. Such challenges drive the need for a comprehensive understanding - not just identifying these issues but delving into their possible implications and existing solutions to build more reliable language models tailored to code intelligence. Based on a well-defined systematic research approach, we conducted an extensive literature review to uncover the pitfalls inherent in LM4Code. Finally, 67 primary studies from top-tier venues have been identified. After carefully examining these studies, we designed a taxonomy of pitfalls in LM4Code research and conducted a systematic study to summarize the issues, implications, current solutions, and challenges of different pitfalls for LM4Code systems. We developed a comprehensive classification scheme that dissects pitfalls across four crucial aspects: data collection and labeling, system design and learning, performance evaluation, and deployment and maintenance. Through this study, we aim to provide a roadmap for researchers and practitioners, facilitating their understanding and utilization of LM4Code in reliable and trustworthy ways.
Neural language models are probabilistic models of human text. They are predominantly trained using maximum likelihood estimation (MLE), which is equivalent to minimizing the forward cross-entropy between the empirical data distribution and the model distribution. However, various degeneration phenomena are still widely observed when decoding from the distributions learned by such models. We establish that the forward cross-entropy is suboptimal as a distance metric for aligning human and model distribution due to its (1) recall-prioritization (2) negative diversity ignorance and (3) train-test mismatch. In this paper, we propose Earth Mover Distance Optimization (EMO) for auto-regressive language modeling. EMO capitalizes on the inherent properties of earth mover distance to address the aforementioned challenges. Due to the high complexity of direct computation, we further introduce a feasible upper bound for EMO to ease end-to-end training. Upon extensive evaluation of language models trained using EMO and MLE. We find that EMO demonstrates a consistently better language modeling performance than MLE across domains. Moreover, EMO demonstrates noteworthy enhancements in downstream performance with minimal fine-tuning on merely 25,000 sentences. This highlights the tremendous potential of EMO as a lightweight calibration method for enhancing large-scale pre-trained language models.
Backdoor attacks have emerged as a prominent threat to natural language processing (NLP) models, where the presence of specific triggers in the input can lead poisoned models to misclassify these inputs to predetermined target classes. Current detection mechanisms are limited by their inability to address more covert backdoor strategies, such as style-based attacks. In this work, we propose an innovative test-time poisoned sample detection framework that hinges on the interpretability of model predictions, grounded in the semantic meaning of inputs. We contend that triggers (e.g., infrequent words) are not supposed to fundamentally alter the underlying semantic meanings of poisoned samples as they want to stay stealthy. Based on this observation, we hypothesize that while the model's predictions for paraphrased clean samples should remain stable, predictions for poisoned samples should revert to their true labels upon the mutations applied to triggers during the paraphrasing process. We employ ChatGPT, a state-of-the-art large language model, as our paraphraser and formulate the trigger-removal task as a prompt engineering problem. We adopt fuzzing, a technique commonly used for unearthing software vulnerabilities, to discover optimal paraphrase prompts that can effectively eliminate triggers while concurrently maintaining input semantics. Experiments on 4 types of backdoor attacks, including the subtle style backdoors, and 4 distinct datasets demonstrate that our approach surpasses baseline methods, including STRIP, RAP, and ONION, in precision and recall.
While large language models have proven effective in a huge range of downstream applications, they often generate text that is problematic or lacks a desired attribute. In this paper, we introduce Reward-Augmented Decoding (RAD), a text generation procedure that uses a small unidirectional reward model to encourage a language model to generate text that has certain properties. Specifically, RAD uses the reward model to score generations as they are produced and rescales sampling probabilities to favor high-reward tokens. By using a unidirectional reward model, RAD can cache activations from prior generation steps to decrease computational overhead. Through experiments on generating non-toxic and sentiment-controlled text, we demonstrate that RAD performs best among methods that change only the generation procedure and matches the performance of state-of-the-art methods that involve re-training the language model. We further validate that RAD is effective on very large language models while incurring a minimal computational overhead.
Program verifiers for imperative languages such as C may be annotation-based, in which assertions and invariants are put into source files and then checked, or tactic-based, where proof scripts separate from programs are interactively developed in a proof assistant such as Coq. Annotation verifiers have been more automated and convenient, but some interactive verifiers have richer assertion languages and formal proofs of soundness. We present VST-A, an annotation verifier that uses the rich assertion language of VST, leverages the formal soundness proof of VST, but allows users to describe functional correctness proofs intuitively by inserting assertions. VST-A analyzes control flow graphs, decomposes every C function into control flow paths between assertions, and reduces program verification problems into corresponding straightline Hoare triples. Compared to existing foundational program verification tools like VST and Iris, in VST-A, such decompositions and reductions are allowed to be nonstructural, which makes VST-A more flexible to use. VST-A's decomposition and reduction is defined in Coq, proved sound in Coq, and computed in a call-by-value way in Coq. The soundness proof for reduction is totally logical, independent of the complicated semantic model (and soundness proof) of VST's Hoare triple. Because of the rich assertion language, not all reduced proof goals can be automatically checked, but the system allows users to prove residual proof goals using the full power of the Coq proof assistant.
Instruction tuning has emerged to enhance the capabilities of large language models (LLMs) to comprehend instructions and generate appropriate responses. Existing methods either manually annotate or employ LLM (e.g., GPT-series) to generate data for instruction tuning. However, they often overlook associating instructions with existing annotated datasets. In this paper, we propose Dynosaur, a dynamic growth paradigm for the automatic curation of instruction-tuning data. Based on the metadata of existing datasets, we use LLMs to automatically construct instruction-tuning data by identifying relevant data fields and generating appropriate instructions. By leveraging the existing annotated datasets, Dynosaur offers several advantages: 1) it reduces the API cost for generating instructions (e.g., it costs less than $12 USD by calling GPT-3.5-turbo for generating 800K instruction tuning samples; 2) it provides high-quality data for instruction tuning (e.g., it performs better than Alpaca and Flan on Super-NI and Longform with comparable data sizes); and 3) it supports the continuous improvement of models by generating instruction-tuning data when a new annotated dataset becomes available. We further investigate a continual learning scheme for learning with the ever-growing instruction-tuning dataset, and demonstrate that replaying tasks with diverse instruction embeddings not only helps mitigate forgetting issues but generalizes to unseen tasks better. Code and data are available at //github.com/WadeYin9712/Dynosaur.
Interpretability methods are developed to understand the working mechanisms of black-box models, which is crucial to their responsible deployment. Fulfilling this goal requires both that the explanations generated by these methods are correct and that people can easily and reliably understand them. While the former has been addressed in prior work, the latter is often overlooked, resulting in informal model understanding derived from a handful of local explanations. In this paper, we introduce explanation summary (ExSum), a mathematical framework for quantifying model understanding, and propose metrics for its quality assessment. On two domains, ExSum highlights various limitations in the current practice, helps develop accurate model understanding, and reveals easily overlooked properties of the model. We also connect understandability to other properties of explanations such as human alignment, robustness, and counterfactual minimality and plausibility.
Semantic, instance, and panoptic segmentations have been addressed using different and specialized frameworks despite their underlying connections. This paper presents a unified, simple, and effective framework for these essentially similar tasks. The framework, named K-Net, segments both instances and semantic categories consistently by a group of learnable kernels, where each kernel is responsible for generating a mask for either a potential instance or a stuff class. To remedy the difficulties of distinguishing various instances, we propose a kernel update strategy that enables each kernel dynamic and conditional on its meaningful group in the input image. K-Net can be trained in an end-to-end manner with bipartite matching, and its training and inference are naturally NMS-free and box-free. Without bells and whistles, K-Net surpasses all previous published state-of-the-art single-model results of panoptic segmentation on MS COCO test-dev split and semantic segmentation on ADE20K val split with 55.2% PQ and 54.3% mIoU, respectively. Its instance segmentation performance is also on par with Cascade Mask R-CNN on MS COCO with 60%-90% faster inference speeds. Code and models will be released at //github.com/ZwwWayne/K-Net/.
Explainable recommendation attempts to develop models that generate not only high-quality recommendations but also intuitive explanations. The explanations may either be post-hoc or directly come from an explainable model (also called interpretable or transparent model in some context). Explainable recommendation tries to address the problem of why: by providing explanations to users or system designers, it helps humans to understand why certain items are recommended by the algorithm, where the human can either be users or system designers. Explainable recommendation helps to improve the transparency, persuasiveness, effectiveness, trustworthiness, and satisfaction of recommendation systems. In this survey, we review works on explainable recommendation in or before the year of 2019. We first highlight the position of explainable recommendation in recommender system research by categorizing recommendation problems into the 5W, i.e., what, when, who, where, and why. We then conduct a comprehensive survey of explainable recommendation on three perspectives: 1) We provide a chronological research timeline of explainable recommendation, including user study approaches in the early years and more recent model-based approaches. 2) We provide a two-dimensional taxonomy to classify existing explainable recommendation research: one dimension is the information source (or display style) of the explanations, and the other dimension is the algorithmic mechanism to generate explainable recommendations. 3) We summarize how explainable recommendation applies to different recommendation tasks, such as product recommendation, social recommendation, and POI recommendation. We also devote a section to discuss the explanation perspectives in broader IR and AI/ML research. We end the survey by discussing potential future directions to promote the explainable recommendation research area and beyond.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191