Wenhao Liu

Tian Xie, Xinyi Yang, Angela S. Lin et al.

Creating a system that can have meaningful conversations with humans to help accomplish tasks is one of the ultimate goals of Artificial Intelligence (AI). It has defined the meaning of AI since the beginning. A lot has been accomplished in this area recently, with voice assistant products entering our daily lives and chat bot systems becoming commonplace in customer service. At first glance there seems to be no shortage of options for dialogue systems. However, the frequently deployed dialogue systems today seem to all struggle with a critical weakness - they are hard to build and harder to maintain. At the core of the struggle is the need to script every single turn of interactions between the bot and the human user. This makes the dialogue systems more difficult to maintain as the tasks become more complex and more tasks are added to the system. In this paper, we propose Converse, a flexible tree-based modular task-oriented dialogue system. Converse uses an and-or tree structure to represent tasks and offers powerful multi-task dialogue management. Converse supports task dependency and task switching, which are unique features compared to other open-source dialogue frameworks. At the same time, Converse aims to make the bot building process easy and simple, for both professional and non-professional software developers. The code is available at https://github.com/salesforce/Converse.

Philippe Laban, Chien-Sheng Wu, Lidiya Murakhovs'ka et al. · microsoft-research, salesforce

Question generation (QGen) models are often evaluated with standardized NLG metrics that are based on n-gram overlap. In this paper, we measure whether these metric improvements translate to gains in a practical setting, focusing on the use case of helping teachers automate the generation of reading comprehension quizzes. In our study, teachers building a quiz receive question suggestions, which they can either accept or refuse with a reason. Even though we find that recent progress in QGen leads to a significant increase in question acceptance rates, there is still large room for improvement, with the best model having only 68.4% of its questions accepted by the ten teachers who participated in our study. We then leverage the annotations we collected to analyze standard NLG metrics and find that model performance has reached projected upper-bounds, suggesting new automatic metrics are needed to guide QGen research forward.

Philippe Laban, Chien-Sheng Wu, Wenhao Liu et al. · microsoft-research, salesforce

Precisely assessing the progress in natural language generation (NLG) tasks is challenging, and human evaluation to establish a preference in a model's output over another is often necessary. However, human evaluation is usually costly, difficult to reproduce, and non-reusable. In this paper, we propose a new and simple automatic evaluation method for NLG called Near-Negative Distinction (NND) that repurposes prior human annotations into NND tests. In an NND test, an NLG model must place a higher likelihood on a high-quality output candidate than on a near-negative candidate with a known error. Model performance is established by the number of NND tests a model passes, as well as the distribution over task-specific errors the model fails on. Through experiments on three NLG tasks (question generation, question answering, and summarization), we show that NND achieves a higher correlation with human judgments than standard NLG evaluation metrics. We then illustrate NND evaluation in four practical scenarios, for example performing fine-grain model analysis, or studying model training dynamics. Our findings suggest that NND can give a second life to human annotations and provide low-cost NLG evaluation.

Lifu Tu, Jin Qu, Semih Yavuz et al.

Cross-lingual transfer of language models trained on high-resource languages like English has been widely studied for many NLP tasks, but focus on conversational tasks has been rather limited. This is partly due to the high cost of obtaining non-English conversational data, which results in limited coverage. In this work, we introduce XSGD for cross-lingual alignment pretraining, a parallel and large-scale multilingual conversation dataset that we created by translating the English-only Schema-Guided Dialogue (SGD) dataset (Rastogi et al., 2020) into 105 other languages. XSGD contains approximately 330k utterances per language. To facilitate aligned cross-lingual representations, we develop an efficient prompt-tuning-based method for learning alignment prompts. We also investigate two different classifiers: NLI-based and vanilla classifiers, and test cross-lingual capability enabled by the aligned prompts. We evaluate our model's cross-lingual generalization capabilities on two conversation tasks: slot-filling and intent classification. Our results demonstrate the strong and efficient modeling ability of NLI-based classifiers and the large cross-lingual transfer improvements achieved by our aligned prompts, particularly in few-shot settings. In addition, we highlight the nice results of our approach compared to LLMs such as text-davinci-003 and ChatGPT in both zero-shot and few-shot settings. While LLMs exhibit impressive performance in English, their cross-lingual capabilities in other languages, particularly low-resource languages, are limited.

Ehsan Hosseini-Asl, Wenhao Liu, Caiming Xiong

Sentiment analysis is an important task in natural language processing. In recent works, pre-trained language models are often used to achieve state-of-the-art results, especially when training data is scarce. It is common to fine-tune on the downstream task, usually by adding task-specific layers on top of the model. In this paper, we focus on aspect-based sentiment analysis, which involves extracting aspect term, category, and predicting their corresponding polarities. In particular, we are interested in few-shot settings. We propose to reformulate the extraction and prediction tasks into the sequence generation task, using a generative language model with unidirectional attention (GPT2 is used unless stated otherwise). This way, the model learns to accomplish the tasks via language generation without the need of training task-specific layers. Our evaluation results on the single-task polarity prediction show that our approach outperforms the previous state-of-the-art (based on BERT) on average performance by a large margins in few-shot and full-shot settings. More importantly, our generative approach significantly reduces the model variance caused by low-resource data. We further demonstrate that the proposed generative language model can handle joint and multi-task settings, unlike previous work. We observe that the proposed sequence generation method achieves further improved performances on polarity prediction when the model is trained via joint and multi-task settings. Further evaluation on similar sentiment analysis datasets, SST-2, SST- and OOS intent detection validates the superiority and noise robustness of generative language model in few-shot settings.

Prafulla Kumar Choubey, Yu Bai, Chien-Sheng Wu et al. · salesforce

Pre-trained language models (PLMs) have been shown effective for zero-shot (0shot) text classification. 0shot models based on natural language inference (NLI) and next sentence prediction (NSP) employ cross-encoder architecture and infer by making a forward pass through the model for each label-text pair separately. This increases the computational cost to make inferences linearly in the number of labels. In this work, we improve the efficiency of such cross-encoder-based 0shot models by restricting the number of likely labels using another fast base classifier-based conformal predictor (CP) calibrated on samples labeled by the 0shot model. Since a CP generates prediction sets with coverage guarantees, it reduces the number of target labels without excluding the most probable label based on the 0shot model. We experiment with three intent and two topic classification datasets. With a suitable CP for each dataset, we reduce the average inference time for NLI- and NSP-based models by 25.6% and 22.2% respectively, without dropping performance below the predefined error rate of 1%.

Changze Lv, Tianlong Li, Wenhao Liu et al.

Spiking Neural Networks (SNNs) have emerged as a promising alternative to conventional Artificial Neural Networks (ANNs), demonstrating comparable performance in both visual and linguistic tasks while offering the advantage of improved energy efficiency. Despite these advancements, the integration of linguistic and visual features into a unified representation through spike trains poses a significant challenge, and the application of SNNs to multimodal scenarios remains largely unexplored. This paper presents SpikeCLIP, a novel framework designed to bridge the modality gap in spike-based computation. Our approach employs a two-step recipe: an ``alignment pre-training'' to align features across modalities, followed by a ``dual-loss fine-tuning'' to refine the model's performance. Extensive experiments reveal that SNNs achieve results on par with ANNs while substantially reducing energy consumption across various datasets commonly used for multimodal model evaluation. Furthermore, SpikeCLIP maintains robust image classification capabilities, even when dealing with classes that fall outside predefined categories. This study marks a significant advancement in the development of energy-efficient and biologically plausible multimodal learning systems. Our code is available at https://github.com/Lvchangze/SpikeCLIP.

Xiaohua Wang, Muzhao Tian, Yuqi Zeng et al.

Reinforcement Learning from Human Feedback (RLHF) and related alignment paradigms have become central to steering large language models (LLMs) and multimodal large language models (MLLMs) toward human-preferred behaviors. However, these approaches introduce a systemic vulnerability: reward hacking, where models exploit imperfections in learned reward signals to maximize proxy objectives without fulfilling true task intent. As models scale and optimization intensifies, such exploitation manifests as verbosity bias, sycophancy, hallucinated justification, benchmark overfitting, and, in multimodal settings, perception--reasoning decoupling and evaluator manipulation. Recent evidence further suggests that seemingly benign shortcut behaviors can generalize into broader forms of misalignment, including deception and strategic gaming of oversight mechanisms. In this survey, we propose the Proxy Compression Hypothesis (PCH) as a unifying framework for understanding reward hacking. We formalize reward hacking as an emergent consequence of optimizing expressive policies against compressed reward representations of high-dimensional human objectives. Under this view, reward hacking arises from the interaction of objective compression, optimization amplification, and evaluator--policy co-adaptation. This perspective unifies empirical phenomena across RLHF, RLAIF, and RLVR regimes, and explains how local shortcut learning can generalize into broader forms of misalignment, including deception and strategic manipulation of oversight mechanisms. We further organize detection and mitigation strategies according to how they intervene on compression, amplification, or co-adaptation dynamics. By framing reward hacking as a structural instability of proxy-based alignment under scale, we highlight open challenges in scalable oversight, multimodal grounding, and agentic autonomy.

Tianlong Li, Wenhao Liu, Muling Wu et al.

Personality is a crucial factor that shapes human communication patterns, thereby regulating the personalities of large language models (LLMs) holds significant potential in enhancing their user experiences. Previous approaches either relied on fine-tuning LLMs on specific corpora or required manually crafted prompts to evoke specific personalities from LLMs. However, the former is inefficient and costly, while the latter cannot precisely manipulate personality traits at a fine-grained level. To address these challenges, we propose UPLex, a method that uses an Unsupervisedly-Built Personalized Lexicon (UPL) during the decoding phase to manipulate LLM's personality traits. UPL can be constructed from a newly built situational judgment test dataset in an unsupervised fashion, and used to modulate the personality expression of LLMs by dynamically altering their predicted probability of upcoming words in a pluggable fashion. Extensive experimentation demonstrates the remarkable effectiveness and pluggability of our method for fine-grained manipulation of LLMs' personalities.

Tianlong Li, Zhenghua Wang, Wenhao Liu et al.

The recent surge in jailbreaking attacks has revealed significant vulnerabilities in Large Language Models (LLMs) when exposed to malicious inputs. While various defense strategies have been proposed to mitigate these threats, there has been limited research into the underlying mechanisms that make LLMs vulnerable to such attacks. In this study, we suggest that the self-safeguarding capability of LLMs is linked to specific activity patterns within their representation space. Although these patterns have little impact on the semantic content of the generated text, they play a crucial role in shaping LLM behavior under jailbreaking attacks. Our findings demonstrate that these patterns can be detected with just a few pairs of contrastive queries. Extensive experimentation shows that the robustness of LLMs against jailbreaking can be manipulated by weakening or strengthening these patterns. Further visual analysis provides additional evidence for our conclusions, providing new insights into the jailbreaking phenomenon. These findings highlight the importance of addressing the potential misuse of open-source LLMs within the community.

Haoran Luo, Haihong E, Yikai Guo et al. · mit

Knowledge Base Question Answering (KBQA) aims to answer natural language questions with a large-scale structured knowledge base (KB). Despite advancements with large language models (LLMs), KBQA still faces challenges in weak KB awareness, imbalance between effectiveness and efficiency, and high reliance on annotated data. To address these challenges, we propose KBQA-o1, a novel agentic KBQA method with Monte Carlo Tree Search (MCTS). It introduces a ReAct-based agent process for stepwise logical form generation with KB environment exploration. Moreover, it employs MCTS, a heuristic search method driven by policy and reward models, to balance agentic exploration's performance and search space. With heuristic exploration, KBQA-o1 generates high-quality annotations for further improvement by incremental fine-tuning. Experimental results show that KBQA-o1 outperforms previous low-resource KBQA methods with limited annotated data, boosting Llama-3.1-8B model's GrailQA F1 performance to 78.5% compared to 48.5% of the previous sota method with GPT-3.5-turbo. Our code is publicly available.

Wenhao Liu, Siyu An, Junru Lu et al.

Role-Playing Agents (RPAs) have shown remarkable performance in various applications, yet they often struggle to recognize and appropriately respond to hard queries that conflict with their role-play knowledge. To investigate RPAs' performance when faced with different types of conflicting requests, we develop an evaluation benchmark that includes contextual knowledge conflicting requests, parametric knowledge conflicting requests, and non-conflicting requests to assess RPAs' ability to identify conflicts and refuse to answer appropriately without over-refusing. Through extensive evaluation, we find that most RPAs behave significant performance gaps toward different conflict requests. To elucidate the reasons, we conduct an in-depth representation-level analysis of RPAs under various conflict scenarios. Our findings reveal the existence of rejection regions and direct response regions within the model's forwarding representation, and thus influence the RPA's final response behavior. Therefore, we introduce a lightweight representation editing approach that conveniently shifts conflicting requests to the rejection region, thereby enhancing the model's refusal accuracy. The experimental results validate the effectiveness of our editing method, improving RPAs' refusal ability of conflicting requests while maintaining their general role-playing capabilities.

Qi Qian, Chengsong Huang, Jingwen Xu et al.

The rapid proliferation of benchmarks for evaluating large language models (LLMs) has created an urgent need for systematic methods to assess benchmark quality itself. We propose Benchmark^2, a comprehensive framework comprising three complementary metrics: (1) Cross-Benchmark Ranking Consistency, measuring whether a benchmark produces model rankings aligned with peer benchmarks; (2) Discriminability Score, quantifying a benchmark's ability to differentiate between models; and (3) Capability Alignment Deviation, identifying problematic instances where stronger models fail but weaker models succeed within the same model family. We conduct extensive experiments across 15 benchmarks spanning mathematics, reasoning, and knowledge domains, evaluating 11 LLMs across four model families. Our analysis reveals significant quality variations among existing benchmarks and demonstrates that selective benchmark construction based on our metrics can achieve comparable evaluation performance with substantially reduced test sets.

Yuanzhe Shen, Zisu Huang, Zhengyuan Wang et al.

As LLM-based agents are deployed in increasingly complex real-world settings, existing benchmarks underrepresent key challenges such as enforcing global constraints, coordinating multi-tool reasoning, and adapting to evolving user behavior over long, multi-turn interactions. To bridge this gap, we introduce \textbf{TRIP-Bench}, a long-horizon benchmark grounded in realistic travel-planning scenarios. TRIP-Bench leverages real-world data, offers 18 curated tools and 40+ travel requirements, and supports automated evaluation. It includes splits of varying difficulty; the hard split emphasizes long and ambiguous interactions, style shifts, feasibility changes, and iterative version revision. Dialogues span up to 15 user turns, can involve 150+ tool calls, and may exceed 200k tokens of context. Experiments show that even advanced models achieve at most 50\% success on the easy split, with performance dropping below 10\% on hard subsets. We further propose \textbf{GTPO}, an online multi-turn reinforcement learning method with specialized reward normalization and reward differencing. Applied to Qwen2.5-32B-Instruct, GTPO improves constraint satisfaction and interaction robustness, outperforming Gemini-3-Pro in our evaluation. We expect TRIP-Bench to advance practical long-horizon interactive agents, and GTPO to provide an effective online RL recipe for robust long-horizon training.

Wenhao Liu, Zhenyi Lu, Xinyu Hu et al.

High-quality math datasets are crucial for advancing the reasoning abilities of large language models (LLMs). However, existing datasets often suffer from three key issues: outdated and insufficient challenging content, neglecting human-like reasoning, and limited reliability due to single-LLM generation. To address these, we introduce STORM-BORN, an ultra-challenging dataset of mathematical derivations sourced from cutting-edge academic papers, which includes dense human-like approximations and heuristic cues. To ensure the reliability and quality, we propose a novel human-in-the-loop, multi-agent data generation framework, integrating reasoning-dense filters, multi-agent collaboration, and human mathematicians' evaluations. We curated a set of 2,000 synthetic samples and deliberately selected the 100 most difficult problems. Even most advanced models like GPT-o1 solved fewer than 5% of them. Fine-tuning on STORM-BORN boosts accuracy by 7.84% (LLaMA3-8B) and 9.12% (Qwen2.5-7B). As AI approaches mathematician-level reasoning, STORM-BORN provides both a high-difficulty benchmark and a human-like reasoning training resource. Our code and dataset are publicly available at https://github.com/lwhere/STORM-BORN.

Zhengyi Jiang, Wenhao Liu, Zhongyi Huang et al.

Analog error-correcting codes (Analog ECCs) for approximate vector-matrix multiplication have been extensively studied as means to achieve fault-tolerant in-memory computation. The theoretical foundations for such coding schemes, particularly the characterization of their correction capabilities via the height profile, have been well established in recent literature. In this paper, we focus on the case of single-error detection Analog ECCs. Among several open problems related to this case proposed by Ron M. Roth in [1], Problem 1 asks: "Identify the values of $k$ and $n$ for which every linear $[n, k]$ code $\mathcal{C}$ over $\mathbb{R}$ satisfies: $$\mathsf{h}_1(\mathcal{C}):=\max_{\boldsymbol{c}\in \mathcal{C}\setminus{\{\boldsymbol{0}\}}}\mathsf{h}_1(\boldsymbol{c})\geq \Big\lceil \frac{k}{n-k} \Big\rceil.\text{"}$$ Here, for any $\boldsymbol{x}\in\mathbb{R}^n$, $\mathsf{h}_1(\boldsymbol{x})$ represents the ratio between the largest and second largest absolute values of $\boldsymbol{x}$'s entries. As the simplest special case of Problem 1 (with $n-k=2$), the following problem was posed as Problem 2 in [1]: "Must every $(n-2)$-dimensional subspace of $\mathbb{R}^n$, $n$ even, contain a nonzero vector in which the ratio between the largest and second largest absolute values of its entries is at least $(n/2)-1$?" These problems directly pertain to the lower bounds on the single-error detection threshold for Analog ECCs: Problem 1 corresponds to arbitrary $n-k$ and Problem 2 corresponds to $n-k=2$. In this paper, we provide an affirmative answer to Problem 2 and a rigorous proof using theories related to convex optimization. Furthermore, we extend our analytical method to show that the lower bound in Problem 1 is tight for the case where $n-k$ divides $k$. Our results fill the gap in the lower bound theory of thresholds for single-error detection in Analog ECCs.

Jesse Vig, Alexander R. Fabbri, Wojciech Kryściński et al.

Query-focused summarization (QFS) aims to produce summaries that answer particular questions of interest, enabling greater user control and personalization. While recently released datasets, such as QMSum or AQuaMuSe, facilitate research efforts in QFS, the field lacks a comprehensive study of the broad space of applicable modeling methods. In this paper we conduct a systematic exploration of neural approaches to QFS, considering two general classes of methods: two-stage extractive-abstractive solutions and end-to-end models. Within those categories, we investigate existing models and explore strategies for transfer learning. We also present two modeling extensions that achieve state-of-the-art performance on the QMSum dataset, up to a margin of 3.38 ROUGE-1, 3.72 ROUGE2, and 3.28 ROUGE-L when combined with transfer learning strategies. Results from human evaluation suggest that the best models produce more comprehensive and factually consistent summaries compared to a baseline model. Code and checkpoints are made publicly available: https://github.com/salesforce/query-focused-sum.

Mingfei Gao, Chen Xing, Juan Carlos Niebles et al.

Despite great progress in object detection, most existing methods work only on a limited set of object categories, due to the tremendous human effort needed for bounding-box annotations of training data. To alleviate the problem, recent open vocabulary and zero-shot detection methods attempt to detect novel object categories beyond those seen during training. They achieve this goal by training on a pre-defined base categories to induce generalization to novel objects. However, their potential is still constrained by the small set of base categories available for training. To enlarge the set of base classes, we propose a method to automatically generate pseudo bounding-box annotations of diverse objects from large-scale image-caption pairs. Our method leverages the localization ability of pre-trained vision-language models to generate pseudo bounding-box labels and then directly uses them for training object detectors. Experimental results show that our method outperforms the state-of-the-art open vocabulary detector by 8% AP on COCO novel categories, by 6.3% AP on PASCAL VOC, by 2.3% AP on Objects365 and by 2.8% AP on LVIS. Code is available at https://github.com/salesforce/PB-OVD.

Wenhao Liu, Xiaohua Wang, Muling Wu et al.

Aligning large language models (LLMs) with human preferences is crucial for enhancing their utility in terms of helpfulness, truthfulness, safety, harmlessness, and interestingness. Existing methods for achieving this alignment often involves employing reinforcement learning from human feedback (RLHF) to fine-tune LLMs based on human labels assessing the relative quality of model responses. Nevertheless, RLHF is susceptible to instability during fine-tuning and presents challenges in implementation.Drawing inspiration from the emerging field of representation engineering (RepE), this study aims to identify relevant representations for high-level human preferences embedded in patterns of activity within an LLM, and achieve precise control of model behavior by transforming its representations. This novel approach, denoted as Representation Alignment from Human Feedback (RAHF), proves to be effective, computationally efficient, and easy to implement.Extensive experiments demonstrate the efficacy of RAHF in not only capturing but also manipulating representations to align with a broad spectrum of human preferences or values, rather than being confined to a singular concept or function (e.g. honesty or bias). RAHF's versatility in accommodating diverse human preferences shows its potential for advancing LLM performance.

Muling Wu, Wenhao Liu, Xiaohua Wang et al.

Parameter Efficient Fine-Tuning (PEFT) techniques have drawn significant attention due to their ability to yield competitive results while updating only a small portion of the adjustable parameters. However, existing PEFT methods pose challenges in hyperparameter selection, such as choosing the rank for LoRA or Adapter, or specifying the length of soft prompts. To address these challenges, we propose a novel fine-tuning approach for neural models, named Representation EDiting (RED), which modifies the representations generated at some layers through the application of scaling and biasing operations. While existing PEFT methods still demonstrate over-parameterization that could potentially undermine the generalization ability acquired from pre-training, RED can substantially reduce the number of trainable parameters by a factor of 25, 700 compared to full parameter fine-tuning and by a factor of 32 relative to LoRA. Remarkably, RED achieves results comparable or superior to both full parameter fine-tuning and other PEFT methods. Extensive experiments across various model architectures and scales, including RoBERTa, GPT-2, T5, and LLaMA-2, have demonstrated the effectiveness and efficiency of RED1, thereby positioning it as a promising PEFT strategy for large-scale neural models.

Ping Xu, Zhiyuan Ning, Pengjiang Li et al.

Single-cell RNA sequencing (scRNA-seq) reveals cell heterogeneity, with cell clustering playing a key role in identifying cell types and marker genes. Recent advances, especially graph neural networks (GNNs)-based methods, have significantly improved clustering performance. However, the analysis of scRNA-seq data remains challenging due to noise, sparsity, and high dimensionality. Compounding these challenges, GNNs often suffer from over-smoothing, limiting their ability to capture complex biological information. In response, we propose scSiameseClu, a novel Siamese Clustering framework for interpreting single-cell RNA-seq data, comprising of 3 key steps: (1) Dual Augmentation Module, which applies biologically informed perturbations to the gene expression matrix and cell graph relationships to enhance representation robustness; (2) Siamese Fusion Module, which combines cross-correlation refinement and adaptive information fusion to capture complex cellular relationships while mitigating over-smoothing; and (3) Optimal Transport Clustering, which utilizes Sinkhorn distance to efficiently align cluster assignments with predefined proportions while maintaining balance. Comprehensive evaluations on seven real-world datasets demonstrate that scSiameseClu outperforms state-of-the-art methods in single-cell clustering, cell type annotation, and cell type classification, providing a powerful tool for scRNA-seq data interpretation.

Muling Wu, Qi Qian, Wenhao Liu et al.

Large Language Models (LLMs) have achieved remarkable performance across various reasoning tasks, yet post-training is constrained by inefficient sample utilization and inflexible difficulty samples processing. To address these limitations, we propose Customized Curriculum Learning (CCL), a novel framework with two key innovations. First, we introduce model-adaptive difficulty definition that customizes curriculum datasets based on each model's individual capabilities rather than using predefined difficulty metrics. Second, we develop "Guided Prompting," which dynamically reduces sample difficulty through strategic hints, enabling effective utilization of challenging samples that would otherwise degrade performance. Comprehensive experiments on supervised fine-tuning and reinforcement learning demonstrate that CCL significantly outperforms uniform training approaches across five mathematical reasoning benchmarks, confirming its effectiveness across both paradigms in enhancing sample utilization and model performance.

Zhengkang Guo, Wenhao Liu, Mingchen Xie et al.

Large language models (LLMs) are increasingly expected to tackle complex tasks, driven by their expanding applications and users' growing proficiency in crafting sophisticated prompts. However, as the number of explicitly stated requirements increases (particularly more than 10 constraints), LLMs often struggle to accurately follow such complex instructions, which limits their applicability in complex real-world scenarios. To the best of our knowledge, existing datasets do not exceed 10 constraints per instance. To address this challenge, we propose RECAST, an efficient and scalable framework for synthesizing datasets where each example incorporates far more constraints than those in existing benchmarks, aiming to challenge and extend the boundaries of models' ability to follow complex instructions. These constraints are extracted from real-world prompt-response pairs to ensure practical relevance. Using this framework, we construct RECAST-30K, a large-scale, high-quality dataset comprising 30k instances spanning 19 constraint types. Experimental results demonstrate that models finetuned on RECAST-30K substantially improve in following complex instructions while maintaining their general capabilities without degradation. Moreover, RECAST enables automatic verification of constraint satisfaction via rule-based validators for quantitative constraints and LLM-based validators for qualitative ones; the verifiability provided by RECAST enables the design of reward functions for reinforcement learning, which further boosts model performance on complex and challenging tasks.

JianHao Zhu, Changze Lv, Xiaohua Wang et al.

Conventional federated learning primarily aims to secure the privacy of data distributed across multiple edge devices, with the global model dispatched to edge devices for parameter updates during the learning process. However, the development of large language models (LLMs) requires substantial data and computational resources, rendering them valuable intellectual properties for their developers and owners. To establish a mechanism that protects both data and model privacy in a federated learning context, we introduce a method that just needs to distribute a quantized version of the model's parameters during training. This method enables accurate gradient estimations for parameter updates while preventing clients from accessing a model whose performance is comparable to the centrally hosted one. Moreover, we combine this quantization strategy with LoRA, a popular and parameter-efficient fine-tuning method, to significantly reduce communication costs in federated learning. The proposed framework, named \textsc{FedLPP}, successfully ensures both data and model privacy in the federated learning context. Additionally, the learned central model exhibits good generalization and can be trained in a resource-efficient manner.

Liang Qiu, Chien-Sheng Wu, Wenhao Liu et al.

Extracting structure information from dialogue data can help us better understand user and system behaviors. In task-oriented dialogues, dialogue structure has often been considered as transition graphs among dialogue states. However, annotating dialogue states manually is expensive and time-consuming. In this paper, we propose a simple yet effective approach for structure extraction in task-oriented dialogues. We first detect and cluster possible slot tokens with a pre-trained model to approximate dialogue ontology for a target domain. Then we track the status of each identified token group and derive a state transition structure. Empirical results show that our approach outperforms unsupervised baseline models by far in dialogue structure extraction. In addition, we show that data augmentation based on extracted structures enriches the surface formats of training data and can achieve a significant performance boost in dialogue response generation.

Alexander R. Fabbri, Chien-Sheng Wu, Wenhao Liu et al.

Factual consistency is an essential quality of text summarization models in practical settings. Existing work in evaluating this dimension can be broadly categorized into two lines of research, entailment-based and question answering (QA)-based metrics, and different experimental setups often lead to contrasting conclusions as to which paradigm performs the best. In this work, we conduct an extensive comparison of entailment and QA-based metrics, demonstrating that carefully choosing the components of a QA-based metric, especially question generation and answerability classification, is critical to performance. Building on those insights, we propose an optimized metric, which we call QAFactEval, that leads to a 14% average improvement over previous QA-based metrics on the SummaC factual consistency benchmark, and also outperforms the best-performing entailment-based metric. Moreover, we find that QA-based and entailment-based metrics can offer complementary signals and be combined into a single metric for a further performance boost.

Prakhar Gupta, Chien-Sheng Wu, Wenhao Liu et al.

Fact-checking is an essential tool to mitigate the spread of misinformation and disinformation. We introduce the task of fact-checking in dialogue, which is a relatively unexplored area. We construct DialFact, a testing benchmark dataset of 22,245 annotated conversational claims, paired with pieces of evidence from Wikipedia. There are three sub-tasks in DialFact: 1) Verifiable claim detection task distinguishes whether a response carries verifiable factual information; 2) Evidence retrieval task retrieves the most relevant Wikipedia snippets as evidence; 3) Claim verification task predicts a dialogue response to be supported, refuted, or not enough information. We found that existing fact-checking models trained on non-dialogue data like FEVER fail to perform well on our task, and thus, we propose a simple yet data-efficient solution to effectively improve fact-checking performance in dialogue. We point out unique challenges in DialFact such as handling the colloquialisms, coreferences and retrieval ambiguities in the error analysis to shed light on future research in this direction.

Lidiya Murakhovs'ka, Chien-Sheng Wu, Philippe Laban et al.

Asking good questions is an essential ability for both human and machine intelligence. However, existing neural question generation approaches mainly focus on the short factoid type of answers. In this paper, we propose a neural question generator, MixQG, to bridge this gap. We combine 9 question answering datasets with diverse answer types, including yes/no, multiple-choice, extractive, and abstractive answers, to train a single generative model. We show with empirical results that our model outperforms existing work in both seen and unseen domains and can generate questions with different cognitive levels when conditioned on different answer types. Our code is released and well-integrated with the Huggingface library to facilitate various downstream applications.

Prafulla Kumar Choubey, Alexander R. Fabbri, Jesse Vig et al.

Hallucination is a known issue for neural abstractive summarization models. Recent work suggests that the degree of hallucination may depend on errors in the training data. In this work, we propose a new method called Contrastive Parameter Ensembling (CaPE) to use training data more effectively, utilizing variations in noise in training samples to reduce hallucination. We first select clean and noisy subsets from the training data using different automatic factual metrics. Then, we fine-tune a base summarization model, which is trained on all training samples, on the clean (noisy) subset to obtain an \textit{expert} (\textit{anti-expert}) model. Finally, we adjust the parameters of base model by the difference between parameters of the \textit{expert} and \textit{anti-expert} models, steering the base model towards the \textit{expert} model and away from the \textit{anti-expert} model. Experimental results show that CaPE improves performance across different automatic factual metrics and human evaluation, with the maximum improvement of 16.69\% and 15.78\% on summary-level dependency-arc entailment accuracy for the XSUM and CNN/DM datasets. The improvement in factual performance does not degrade the performance on other metrics of informativeness such as ROUGE.

Zahra Fatemi, Chen Xing, Wenhao Liu et al.

Existing studies addressing gender bias of pre-trained language models, usually build a small gender-neutral data set and conduct a second phase pre-training on the model with such data. However, given the limited size and concentrated focus of the gender-neutral data, catastrophic forgetting would occur during second-phase pre-training. Forgetting information in the original training data may damage the model's downstream performance by a large margin. In this work, we empirically show that catastrophic forgetting occurs in such methods by evaluating them with general NLP tasks in GLUE. Then, we propose a new method, GEnder Equality Prompt (GEEP), to improve gender fairness of pre-trained models with less forgetting. GEEP freezes the pre-trained model and learns gender-related prompts with gender-neutral data. Empirical results show that GEEP not only achieves SOTA performances on gender fairness tasks, but also forgets less and performs better on GLUE by a large margin.

Tanya Goyal, Nazneen Fatema Rajani, Wenhao Liu et al.

Summarization systems make numerous "decisions" about summary properties during inference, e.g. degree of copying, specificity and length of outputs, etc. However, these are implicitly encoded within model parameters and specific styles cannot be enforced. To address this, we introduce HydraSum, a new summarization architecture that extends the single decoder framework of current models to a mixture-of-experts version with multiple decoders. We show that HydraSum's multiple decoders automatically learn contrasting summary styles when trained under the standard training objective without any extra supervision. Through experiments on three summarization datasets (CNN, Newsroom and XSum), we show that HydraSum provides a simple mechanism to obtain stylistically-diverse summaries by sampling from either individual decoders or their mixtures, outperforming baseline models. Finally, we demonstrate that a small modification to the gating strategy during training can enforce an even stricter style partitioning, e.g. high- vs low-abstractiveness or high- vs low-specificity, allowing users to sample from a larger area in the generation space and vary summary styles along multiple dimensions.

Chien-Sheng Wu, Linqing Liu, Wenhao Liu et al.

In this paper, we aim to improve abstractive dialogue summarization quality and, at the same time, enable granularity control. Our model has two primary components and stages: 1) a two-stage generation strategy that generates a preliminary summary sketch serving as the basis for the final summary. This summary sketch provides a weakly supervised signal in the form of pseudo-labeled interrogative pronoun categories and key phrases extracted using a constituency parser. 2) A simple strategy to control the granularity of the final summary, in that our model can automatically determine or control the number of generated summary sentences for a given dialogue by predicting and highlighting different text spans from the source text. Our model achieves state-of-the-art performance on the largest dialogue summarization corpus SAMSum, with as high as 50.79 in ROUGE-L score. In addition, we conduct a case study and show competitive human evaluation results and controllability to human-annotated summaries.

Chien-Sheng Wu, Andrea Madotto, Wenhao Liu et al.

This paper introduces QAConv, a new question answering (QA) dataset that uses conversations as a knowledge source. We focus on informative conversations, including business emails, panel discussions, and work channels. Unlike open-domain and task-oriented dialogues, these conversations are usually long, complex, asynchronous, and involve strong domain knowledge. In total, we collect 34,608 QA pairs from 10,259 selected conversations with both human-written and machine-generated questions. We use a question generator and a dialogue summarizer as auxiliary tools to collect and recommend questions. The dataset has two testing scenarios: chunk mode and full mode, depending on whether the grounded partial conversation is provided or retrieved. Experimental results show that state-of-the-art pretrained QA systems have limited zero-shot performance and tend to predict our questions as unanswerable. Our dataset provides a new training and evaluation testbed to facilitate QA on conversations research.

Chen Xing, Wenhao Liu, Caiming Xiong

Fitting complex patterns in the training data, such as reasoning and commonsense, is a key challenge for language pre-training. According to recent studies and our empirical observations, one possible reason is that some easy-to-fit patterns in the training data, such as frequently co-occurring word combinations, dominate and harm pre-training, making it hard for the model to fit more complex information. We argue that mis-predictions can help locate such dominating patterns that harm language understanding. When a mis-prediction occurs, there should be frequently co-occurring patterns with the mis-predicted word fitted by the model that lead to the mis-prediction. If we can add regularization to train the model to rely less on such dominating patterns when a mis-prediction occurs and focus more on the rest more subtle patterns, more information can be efficiently fitted at pre-training. Following this motivation, we propose a new language pre-training method, Mis-Predictions as Harm Alerts (MPA). In MPA, when a mis-prediction occurs during pre-training, we use its co-occurrence information to guide several heads of the self-attention modules. Some self-attention heads in the Transformer modules are optimized to assign lower attention weights to the words in the input sentence that frequently co-occur with the mis-prediction while assigning higher weights to the other words. By doing so, the Transformer model is trained to rely less on the dominating frequently co-occurring patterns with mis-predictions while focus more on the rest more complex information when mis-predictions occur. Our experiments show that MPA expedites the pre-training of BERT and ELECTRA and improves their performances on downstream tasks.

Jian-Guo Zhang, Kazuma Hashimoto, Wenhao Liu et al.

Intent detection is one of the core components of goal-oriented dialog systems, and detecting out-of-scope (OOS) intents is also a practically important skill. Few-shot learning is attracting much attention to mitigate data scarcity, but OOS detection becomes even more challenging. In this paper, we present a simple yet effective approach, discriminative nearest neighbor classification with deep self-attention. Unlike softmax classifiers, we leverage BERT-style pairwise encoding to train a binary classifier that estimates the best matched training example for a user input. We propose to boost the discriminative ability by transferring a natural language inference (NLI) model. Our extensive experiments on a large-scale multi-domain intent detection task show that our method achieves more stable and accurate in-domain and OOS detection accuracy than RoBERTa-based classifiers and embedding-based nearest neighbor approaches. More notably, the NLI transfer enables our 10-shot model to perform competitively with 50-shot or even full-shot classifiers, while we can keep the inference time constant by leveraging a faster embedding retrieval model.

Wenhao Liu, Maurizio Adriano Strangio, Shengbao Wang

Efficient certificateless one-pass session key establishment protocols can be constructed from key encapsulation mechanisms (KEMs) by making use of tags and signcryption schemes. The resulting primitives are referred to as Certificateless Signcryption Tag Key Encapsulation Mechanisms (CLSC-TKEMs). In this paper we propose two novel CLSC-TKEM protocols, the first, named LSW-CLSC-TKEM, makes use of the signature scheme of Liu et al., the second, named DKTUTS-CLSC-TKEM, is based on the direct key transport using a timestamp (DKTUTS) protocol first described by Zheng. In order to achieve greater efficiency both schemes are instantiated on elliptic curves without making use of pairings and are therefore good candidates for deployment on resource constrained devices.