Socio-demographic prompting is a commonly employed approach to study cultural biases in LLMs as well as for aligning models to certain cultures. In this paper, we systematically probe four LLMs (Llama 3, Mistral v0.2, GPT-3.5 Turbo and GPT-4) with prompts that are conditioned on culturally sensitive and non-sensitive cues, on datasets that are supposed to be culturally sensitive (EtiCor and CALI) or neutral (MMLU and ETHICS). We observe that all models except GPT-4 show significant variations in their responses on both kinds of datasets for both kinds of prompts, casting doubt on the robustness of the culturally-conditioned prompting as a method for eliciting cultural bias in models or as an alignment strategy. The work also calls rethinking the control experiment design to tease apart the cultural conditioning of responses from "placebo effect", i.e., random perturbations of model responses due to arbitrary tokens in the prompt.
相關內容
This study explores the temporal dynamics of language processing by examining the alignment between word representations from a pre-trained transformer-based language model, and EEG data. Using a Temporal Response Function (TRF) model, we investigate how neural activity corresponds to model representations across different layers, revealing insights into the interaction between artificial language models and brain responses during language comprehension. Our analysis reveals patterns in TRFs from distinct layers, highlighting varying contributions to lexical and compositional processing. Additionally, we used linear discriminant analysis (LDA) to isolate part-of-speech (POS) representations, offering insights into their influence on neural responses and the underlying mechanisms of syntactic processing. These findings underscore EEG's utility for probing language processing dynamics with high temporal resolution. By bridging artificial language models and neural activity, this study advances our understanding of their interaction at fine timescales.
Language model (LM) distillation is a trending area that aims to distil the knowledge residing in a large teacher LM to a small student one. While various methods have been proposed to maximize the effectiveness of the distillation, significant challenges persist, particularly when there is a substantial capacity gap between the teacher and student LMs. This issue, often referred to as the \textit{curse} of capacity gap, suggests that a larger teacher does not necessarily result in a superior student compared to one distilled from a smaller teacher. In other words, there is likely an optimal teacher yielding the best student along the scaling course of the teacher. However, the curse of capacity gap can not be tackled without notable compute overhead, as indicated in previous studies. In the context of large LMs (LLMs), previously viable approaches become much less meaningful, as it is an impossible triangle to distill an expected student from an optimal teacher student with small compute overhead. Fortunately, the impossible triangle can fortunately be possible provided an inducted \textit{law} of capacity gap. In this paper, we take the spirits of scaling law and reveal that the optimal teacher scale almost consistently follows a linear scaling with the student scale across different model architectures and data scales. The law later guides us to distil a 3B student LM (termed \textsc{MiniMA}) from LLaMA2-7B. \textsc{MiniMA} is demonstrated to outperform a wide range of 3B competitors and could even compete with several 7B models.
Steering vectors (SVs) are a new approach to efficiently adjust language model behaviour at inference time by intervening on intermediate model activations. They have shown promise in terms of improving both capabilities and model alignment. However, the reliability and generalisation properties of this approach are unknown. In this work, we rigorously investigate these properties, and show that steering vectors have substantial limitations both in- and out-of-distribution. In-distribution, steerability is highly variable across different inputs. Depending on the concept, spurious biases can substantially contribute to how effective steering is for each input, presenting a challenge for the widespread use of steering vectors. Out-of-distribution, while steering vectors often generalise well, for several concepts they are brittle to reasonable changes in the prompt, resulting in them failing to generalise well. Overall, our findings show that while steering can work well in the right circumstances, there remain many technical difficulties of applying steering vectors to guide models' behaviour at scale.
This paper investigates the reliability of explanations generated by large language models (LLMs) when prompted to explain their previous output. We evaluate two kinds of such self-explanations - extractive and counterfactual - using three state-of-the-art LLMs (2B to 8B parameters) on two different classification tasks (objective and subjective). Our findings reveal, that, while these self-explanations can correlate with human judgement, they do not fully and accurately follow the model's decision process, indicating a gap between perceived and actual model reasoning. We show that this gap can be bridged because prompting LLMs for counterfactual explanations can produce faithful, informative, and easy-to-verify results. These counterfactuals offer a promising alternative to traditional explainability methods (e.g. SHAP, LIME), provided that prompts are tailored to specific tasks and checked for validity.
Different scheduling mechanisms in Time Sensitive Networking (TSN) can be integrated together to design and support complex architectures with enhanced capabilities for mixed critical networks. Integrating Frame Preemption (FP) with Credit-Based Shaper (CBS) and Gate Control List (GCL) opens up different modes and configuration choices resulting in a complex evaluation of several possibilities and their impact on the Quality of Service (QoS). In this paper, we implement and quantify the integration of preemptive CBS with GCL by incorporating FP into the architecture. Our experiments show that the end-to-end delay of Audio Video Bridging (AVB) flows shaped by CBS reduces significantly (up to 40\%) when AVB flows are set to preemptable class. We further show that the jitter of Time Triggered (TT) traffic remains unaffected in "with Hold/Release" mode. Furthermore, we propose to introduce Guardband (GB) in the "without Hold/Release" to reduce the jitter of the TT flow. We compare all the different integration modes, starting with CBS with GCL, extending it further to FP. We evaluate all feasible combinations in both synthetic and realistic scenarios and offer recommendations for practical configuration methods.
Are vision-language models (VLMs) for open-vocabulary perception inherently open-set models because they are trained on internet-scale datasets? We answer this question with a clear no - VLMs introduce closed-set assumptions via their finite query set, making them vulnerable to open-set conditions. We systematically evaluate VLMs for open-set recognition and find they frequently misclassify objects not contained in their query set, leading to alarmingly low precision when tuned for high recall and vice versa. We show that naively increasing the size of the query set to contain more and more classes does not mitigate this problem, but instead causes diminishing task performance and open-set performance. We establish a revised definition of the open-set problem for the age of VLMs, define a new benchmark and evaluation protocol to facilitate standardised evaluation and research in this important area, and evaluate promising baseline approaches based on predictive uncertainty and dedicated negative embeddings on a range of open-vocabulary VLM classifiers and object detectors.
Deep learning-based models are widely deployed in autonomous driving areas, especially the increasingly noticed end-to-end solutions. However, the black-box property of these models raises concerns about their trustworthiness and safety for autonomous driving, and how to debug the causality has become a pressing concern. Despite some existing research on the explainability of autonomous driving, there is currently no systematic solution to help researchers debug and identify the key factors that lead to the final predicted action of end-to-end autonomous driving. In this work, we propose a comprehensive approach to explore and analyze the causality of end-to-end autonomous driving. First, we validate the essential information that the final planning depends on by using controlled variables and counterfactual interventions for qualitative analysis. Then, we quantitatively assess the factors influencing model decisions by visualizing and statistically analyzing the response of key model inputs. Finally, based on the comprehensive study of the multi-factorial end-to-end autonomous driving system, we have developed a strong baseline and a tool for exploring causality in the close-loop simulator CARLA. It leverages the essential input sources to obtain a well-designed model, resulting in highly competitive capabilities. As far as we know, our work is the first to unveil the mystery of end-to-end autonomous driving and turn the black box into a white one. Thorough close-loop experiments demonstrate that our method can be applied to end-to-end autonomous driving solutions for causality debugging. Code will be available at //github.com/bdvisl/DriveInsight.
Recent innovations in artificial intelligence (AI), primarily powered by large language models (LLMs), have transformed how programmers develop and maintain software -- leading to new frontiers in software engineering (SE). The advanced capabilities of LLM-based programming assistants to support software development tasks have led to a rise in the adoption of LLMs in SE. However, little is known about the evidenced-based practices, tools and processes verified by research findings, supported and adopted by AI programming assistants. To this end, our work conducts a preliminary evaluation exploring the beliefs and behaviors of LLM used to support software development tasks. We investigate 17 evidence-based claims posited by empirical SE research across five LLM-based programming assistants. Our findings show that LLM-based programming assistants have ambiguous beliefs regarding research claims, lack credible evidence to support responses, and are incapable of adopting practices demonstrated by empirical SE research to support development tasks. Based on our results, we provide implications for practitioners adopting LLM-based programming assistants in development contexts and shed light on future research directions to enhance the reliability and trustworthiness of LLMs -- aiming to increase awareness and adoption of evidence-based SE research findings in practice.
In pace with developments in the research field of artificial intelligence, knowledge graphs (KGs) have attracted a surge of interest from both academia and industry. As a representation of semantic relations between entities, KGs have proven to be particularly relevant for natural language processing (NLP), experiencing a rapid spread and wide adoption within recent years. Given the increasing amount of research work in this area, several KG-related approaches have been surveyed in the NLP research community. However, a comprehensive study that categorizes established topics and reviews the maturity of individual research streams remains absent to this day. Contributing to closing this gap, we systematically analyzed 507 papers from the literature on KGs in NLP. Our survey encompasses a multifaceted review of tasks, research types, and contributions. As a result, we present a structured overview of the research landscape, provide a taxonomy of tasks, summarize our findings, and highlight directions for future work.
The dominating NLP paradigm of training a strong neural predictor to perform one task on a specific dataset has led to state-of-the-art performance in a variety of applications (eg. sentiment classification, span-prediction based question answering or machine translation). However, it builds upon the assumption that the data distribution is stationary, ie. that the data is sampled from a fixed distribution both at training and test time. This way of training is inconsistent with how we as humans are able to learn from and operate within a constantly changing stream of information. Moreover, it is ill-adapted to real-world use cases where the data distribution is expected to shift over the course of a model's lifetime. The first goal of this thesis is to characterize the different forms this shift can take in the context of natural language processing, and propose benchmarks and evaluation metrics to measure its effect on current deep learning architectures. We then proceed to take steps to mitigate the effect of distributional shift on NLP models. To this end, we develop methods based on parametric reformulations of the distributionally robust optimization framework. Empirically, we demonstrate that these approaches yield more robust models as demonstrated on a selection of realistic problems. In the third and final part of this thesis, we explore ways of efficiently adapting existing models to new domains or tasks. Our contribution to this topic takes inspiration from information geometry to derive a new gradient update rule which alleviate catastrophic forgetting issues during adaptation.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191