Modern language models can imitate complex patterns through few-shot learning, enabling them to complete challenging tasks without fine-tuning. However, imitation can also lead models to reproduce inaccuracies or harmful content if present in the context. We study harmful imitation through the lens of a model's internal representations, and identify two related phenomena: overthinking and false induction heads. The first phenomenon, overthinking, appears when we decode predictions from intermediate layers, given correct vs. incorrect few-shot demonstrations. At early layers, both demonstrations induce similar model behavior, but the behavior diverges sharply at some "critical layer", after which the accuracy given incorrect demonstrations progressively decreases. The second phenomenon, false induction heads, are a possible mechanistic cause of overthinking: these are heads in late layers that attend to and copy false information from previous demonstrations, and whose ablation reduces overthinking. Beyond scientific understanding, our results suggest that studying intermediate model computations could be a promising avenue for understanding and guarding against harmful model behaviors.

Deep neural networks (DNNs) have demonstrated their outperformance in various software systems, but also exhibit misbehavior and even result in irreversible disasters. Therefore, it is crucial to identify the misbehavior of DNN-based software and improve DNNs' quality. Test input prioritization is one of the most appealing ways to guarantee DNNs' quality, which prioritizes test inputs so that more bug-revealing inputs can be identified earlier with limited time and manual labeling efforts. However, the existing prioritization methods are still limited from three aspects: certifiability, effectiveness, and generalizability. To overcome the challenges, we propose CertPri, a test input prioritization technique designed based on a movement cost perspective of test inputs in DNNs' feature space. CertPri differs from previous works in three key aspects: (1) certifiable: it provides a formal robustness guarantee for the movement cost; (2) effective: it leverages formally guaranteed movement costs to identify malicious bug-revealing inputs; and (3) generic: it can be applied to various tasks, data, models, and scenarios. Extensive evaluations across 2 tasks (i.e., classification and regression), 6 data forms, 4 model structures, and 2 scenarios (i.e., white-box and black-box) demonstrate CertPri's superior performance. For instance, it significantly improves 53.97% prioritization effectiveness on average compared with baselines. Its robustness and generalizability are 1.41~2.00 times and 1.33~3.39 times that of baselines on average, respectively.

The growth in social media has exacerbated the threat of fake news to individuals and communities. This draws increasing attention to developing efficient and timely rumor detection methods. The prevailing approaches resort to graph neural networks (GNNs) to exploit the post-propagation patterns of the rumor-spreading process. However, these methods lack inherent interpretation of rumor detection due to the black-box nature of GNNs. Moreover, these methods suffer from less robust results as they employ all the propagation patterns for rumor detection. In this paper, we address the above issues with the proposed Diverse Counterfactual Evidence framework for Rumor Detection (DCE-RD). Our intuition is to exploit the diverse counterfactual evidence of an event graph to serve as multi-view interpretations, which are further aggregated for robust rumor detection results. Specifically, our method first designs a subgraph generation strategy to efficiently generate different subgraphs of the event graph. We constrain the removal of these subgraphs to cause the change in rumor detection results. Thus, these subgraphs naturally serve as counterfactual evidence for rumor detection. To achieve multi-view interpretation, we design a diversity loss inspired by Determinantal Point Processes (DPP) to encourage diversity among the counterfactual evidence. A GNN-based rumor detection model further aggregates the diverse counterfactual evidence discovered by the proposed DCE-RD to achieve interpretable and robust rumor detection results. Extensive experiments on two real-world datasets show the superior performance of our method. Our code is available at https://github.com/Vicinity111/DCE-RD.

As an efficient distributed machine learning approach, Federated learning (FL) can obtain a shared model by iterative local model training at the user side and global model aggregating at the central server side, thereby protecting privacy of users. Mobile users in FL systems typically communicate with base stations (BSs) via wireless channels, where training performance could be degraded due to unreliable access caused by user mobility. However, existing work only investigates a static scenario or random initialization of user locations, which fail to capture mobility in real-world networks. To tackle this issue, we propose a practical model for user mobility in FL across multiple BSs, and develop a user scheduling and resource allocation method to minimize the training delay with constrained communication resources. Specifically, we first formulate an optimization problem with user mobility that jointly considers user selection, BS assignment to users, and bandwidth allocation to minimize the latency in each communication round. This optimization problem turned out to be NP-hard and we proposed a delay-aware greedy search algorithm (DAGSA) to solve it. Simulation results show that the proposed algorithm achieves better performance than the state-of-the-art baselines and a certain level of user mobility could improve training performance.

Dialogue-based human-AI collaboration can revolutionize collaborative problem-solving, creative exploration, and social support. To realize this goal, the development of automated agents proficient in skills such as negotiating, following instructions, establishing common ground, and progressing shared tasks is essential. This survey begins by reviewing the evolution of dialogue management paradigms in collaborative dialogue systems, from traditional handcrafted and information-state based methods to AI planning-inspired approaches. It then shifts focus to contemporary data-driven dialogue management techniques, which seek to transfer deep learning successes from form-filling and open-domain settings to collaborative contexts. The paper proceeds to analyze a selected set of recent works that apply neural approaches to collaborative dialogue management, spotlighting prevailing trends in the field. This survey hopes to provide foundational background for future advancements in collaborative dialogue management, particularly as the dialogue systems community continues to embrace the potential of large language models.

This study aims to explore user acceptance of Autonomous Vehicle (AV) policies with improved text-mining methods. Recently, South Korean policymakers have viewed Autonomous Driving Car (ADC) and Autonomous Driving Robot (ADR) as next-generation means of transportation that will reduce the cost of transporting passengers and goods. They support the construction of V2I and V2V communication infrastructures for ADC and recognize that ADR is equivalent to pedestrians to promote its deployment into sidewalks. To fill the gap where end-user acceptance of these policies is not well considered, this study applied two text-mining methods to the comments of graduate students in the fields of Industrial, Mechanical, and Electronics-Electrical-Computer. One is the Co-occurrence Network Analysis (CNA) based on TF-IWF and Dice coefficient, and the other is the Contextual Semantic Network Analysis (C-SNA) based on both KeyBERT, which extracts keywords that contextually represent the comments, and double cosine similarity. The reason for comparing these approaches is to balance interest not only in the implications for the AV policies but also in the need to apply quality text mining to this research domain. Significantly, the limitation of frequency-based text mining, which does not reflect textual context, and the trade-off of adjusting thresholds in Semantic Network Analysis (SNA) were considered. As the results of comparing the two approaches, the C-SNA provided the information necessary to understand users' voices using fewer nodes and features than the CNA. The users who pre-emptively understood the AV policies based on their engineering literacy and the given texts revealed potential risks of the AV accident policies. This study adds suggestions to manage these risks to support the successful deployment of AVs on public roads.

Privacy concerns have attracted increasing attention in data-driven products due to the tendency of machine learning models to memorize sensitive training data. Generating synthetic versions of such data with a formal privacy guarantee, such as differential privacy (DP), provides a promising path to mitigating these privacy concerns, but previous approaches in this direction have typically failed to produce synthetic data of high quality. In this work, we show that a simple and practical recipe in the text domain is effective: simply fine-tuning a pretrained generative language model with DP enables the model to generate useful synthetic text with strong privacy protection. Through extensive empirical analyses on both benchmark and private customer data, we demonstrate that our method produces synthetic text that is competitive in terms of utility with its non-private counterpart, meanwhile providing strong protection against potential privacy leakages.

Text generative models trained via Maximum Likelihood Estimation (MLE) suffer from the notorious exposure bias problem, and Generative Adversarial Networks (GANs) are shown to have potential to tackle this problem. Existing language GANs adopt estimators like REINFORCE or continuous relaxations to model word probabilities. The inherent limitations of such estimators lead current models to rely on pre-training techniques (MLE pre-training or pre-trained embeddings). Representation modeling methods which are free from those limitations, however, are seldomly explored because of their poor performance in previous attempts. Our analyses reveal that invalid sampling methods and unhealthy gradients are the main contributors to such unsatisfactory performance. In this work, we present two techniques to tackle these problems: dropout sampling and fully normalized LSTM. Based on these two techniques, we propose InitialGAN whose parameters are randomly initialized in full. Besides, we introduce a new evaluation metric, Least Coverage Rate, to better evaluate the quality of generated samples. The experimental results demonstrate that InitialGAN outperforms both MLE and other compared models. To the best of our knowledge, it is the first time a language GAN can outperform MLE without using any pre-training techniques.

In recent years, word embeddings have been widely used to measure biases in texts. Even if they have proven to be effective in detecting a wide variety of biases, metrics based on word embeddings lack transparency and interpretability. We analyze an alternative PMI-based metric to quantify biases in texts. It can be expressed as a function of conditional probabilities, which provides a simple interpretation in terms of word co-occurrences. We also prove that it can be approximated by an odds ratio, which allows estimating confidence intervals and statistical significance of textual biases. This approach produces similar results to metrics based on word embeddings when capturing gender gaps of the real world embedded in large corpora.

All day and every day, you constantly receive highly personalized instructions for how to comply with the law, sent directly by your government and law enforcement. You're told how to cross the street, how fast to drive on the way to work, and what you're allowed to say or do online--if you're in any situation that might have legal implications, you're told exactly what to do, in real time. Imagine that the computer system formulating these personal legal directives at mass scale is so complex that no one can explain how it reasons or works. But if you ignore a directive, the system will know, and it'll be used as evidence in the prosecution that's sure to follow. This future may not be far off--automatic detection of lawbreaking is nothing new.