Kaiyan Chang, Yonghao Shi, Chenglong Wang, Hang Zhou, Chi Hu, Xiaoqian Liu, Yingfeng Luo, Yuan Ge, Tong Xiao, JingBo Zhu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recent advances in diffusion language models (DLMs) have presented a promising alternative to traditional autoregressive large language models (LLMs). However, DLMs still lag behind LLMs in reasoning performance, especially as the number of denoising steps decreases. Our analysis reveals that this shortcoming arises primarily from the independent generation of masked tokens across denoising steps, which fails to capture the token correlation. In this paper, we define two types of token correlation: intra-sequence correlation and inter-sequence correlation, and demonstrate that enhancing these correlations improves reasoning performance. To this end, we propose a Multi-Reward Optimization (MRO) approach, which encourages DLMs to consider the token correlation during the denoising process. More specifically, our MRO approach leverages test-time scaling, reject sampling, and reinforcement learning to directly optimize the token correlation with multiple elaborate rewards. Additionally, we introduce group step and importance sampling strategies to mitigate reward variance and enhance sampling efficiency. Through extensive experiments, we demonstrate that MRO not only improves reasoning performance but also achieves significant sampling speedups while maintaining high performance on reasoning benchmarks.

The supervision of novel psychoactive substances (NPSs) is a global problem, and the regulation of NPSs was heavily relied on identifying structural matches in established NPSs databases. However, violators could circumvent legal oversight by altering the side chain structure of recognized NPSs and the existing methods cannot overcome the inaccuracy and lag of supervision. In this study, we propose a scaffold and transformer-based NPS generation and Screening (STNGS) framework to systematically identify and evaluate potential NPSs. A scaffold-based generative model and a rank function with four parts are contained by our framework. Our generative model shows excellent performance in the design and optimization of general molecules and NPS-like molecules by chemical space analysis and property distribution analysis. The rank function includes synthetic accessibility score and frequency score, as well as confidence score and affinity score evaluated by a neural network, which enables the precise positioning of potential NPSs. Applied STNGS framework with molecular docking and a G protein-coupled receptor (GPCR) activation-based sensor (GRAB), we successfully identify three novel synthetic cannabinoids with activity. STNGS constrains the chemical space to generate NPS-like molecules database with diversity and novelty, which assists in the ex-ante regulation of NPSs.

Large Language Models (LLMs) have achieved impressive performance across various reasoning tasks. However, even state-of-the-art LLMs such as ChatGPT are prone to logical errors during their reasoning processes. Existing solutions, such as deploying task-specific verifiers or voting over multiple reasoning paths, either require extensive human annotations or fail in scenarios with inconsistent responses. To address these challenges, we introduce RankPrompt, a new prompting method that enables LLMs to self-rank their responses without additional resources. RankPrompt breaks down the ranking problem into a series of comparisons among diverse responses, leveraging the inherent capabilities of LLMs to generate chains of comparison as contextual exemplars. Our experiments across 11 arithmetic and commonsense reasoning tasks show that RankPrompt significantly enhances the reasoning performance of ChatGPT and GPT-4, with improvements of up to 13%. Moreover, RankPrompt excels in LLM-based automatic evaluations for open-ended tasks, aligning with human judgments 74% of the time in the AlpacaEval dataset. It also exhibits robustness to variations in response order and consistency. Collectively, our results validate RankPrompt as an effective method for eliciting high-quality feedback from language models.

Yuan Ge, Yilun Liu, Chi Hu, Weibin Meng, Shimin Tao, Xiaofeng Zhao, Mahong Xia, Zhang Li, Boxing Chen, Hao Yang, Bei Li, Tong Xiao, JingBo Zhu. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Orthogonal frequency division multiplexing (OFDM) technology has become one of the most popular technologies in modern communications. However, OFDM signals have high peak-to-average power ratio (PAPR), which degrades the system performance. In this paper, a novel PAPR reduction scheme based on Arnold transform is proposed. By using the augmentations and Arnold transform, the proposed scheme can achieve the purpose of getting the satisfactory PAPR performance and the same BER performance as the original OFDM systems. More importantly, there is no side information needed in proposed scheme.

This paper addresses the efficiency challenge of Neural Architecture Search (NAS) by formulating the task as a ranking problem.Previous methods require numerous training examples to estimate the accurate performance of architectures, although the actual goal is to find the distinction between "good" and "bad" candidates.Here we do not resort to performance predictors.Instead, we propose a performance ranking method (RankNAS) via pairwise ranking.It enables efficient architecture search using much fewer training examples.Moreover, we develop an architecture selection method to prune the search space and concentrate on more promising candidates.Extensive experiments on machine translation and language modeling tasks show that RankNAS can design high-performance architectures while being orders of magnitude faster than state-ofthe-art NAS systems.

Neural architecture search (NAS) has advanced significantly in recent years but most NAS systems restrict search to learning architectures of a recurrent or convolutional cell.In this paper, we extend the search space of NAS.In particular, we present a general approach to learn both intra-cell and inter-cell architectures (call it ESS).For a better search result, we design a joint learning method to perform intra-cell and inter-cell NAS simultaneously.We implement our model in a differentiable architecture search system.For recurrent neural language modeling, it outperforms a strong baseline significantly on the PTB and Wiki-Text data, with a new state-of-the-art on PTB.Moreover, the learned architectures show good transferability to other systems.E.g., they improve state-of-the-art systems on the CoNLL and WNUT named entity recognition (NER) tasks and CoNLL chunking task, indicating a promising line of research on large-scale prelearned architectures.

For original partial transmit sequence (O‐PTS), the conjugate operations can be applied to achieve the purpose of ameliorating the peak‐to‐average power ratio (PAPR) performance. But in this way, an observation has been found that some of these candidate signals has the same PAPR value that brings about great redundant computations. Based on this, a low‐complexity PTS is put forward to abate the PAPR of orthogonal frequency division multiplexing (OFDM) systems, thereby avoiding redundant candidate signals. In proposed scheme, a novel scheme employing conjugate operations for generating candidate signals with different PAPR values is given, which avoids the generation of candidate signals with the same PAPR value and saves the needless computations. Meanwhile, the grouping is involved for further reduction in computational complexity. Simulation results illustrate that our scheme has the similar or superior PAPR performance with the remarkable reduction in computational complexity by comparison with O‐PTS.

Robotic swarms are usually designed in a bottom-up way, which can make robotic swarms vulnerable to environmental impact. It is particularly true for the widely used control mode of robotic swarms, where it is often the case that neither the correctness of the swarming tasks at the macro level nor the safety of the interaction among agents at the micro level can be guaranteed. To ensure that the behaviors are safe at runtime, it is necessary to take into account the property guard approaches for robotic swarms in uncertain environments. Runtime enforcement is an approach which can guarantee the given properties in system execution and has no scalability issue. Although some runtime enforcement methods have been studied and applied in different domains, they cannot effectively solve the problem of property enforcement on robotic swarm tasks at present. In this paper, an enforcement method is proposed on swarms which should satisfy multi-level properties in uncertain environments. We introduce a macro-micro property enforcing framework with the notion of agent shields and a discrete-time enforcing mechanism called D -time enforcing. To realize this method, a domain specification language and the corresponding enforcer synthesis algorithms are developed. We then apply the approach to enforce the properties of the simulated robotic swarm in the robotflocksim platform. We evaluate and show the effectiveness of the method with experiments on specific unmanned aerial vehicle swarm tasks.

Various robots are playing critical roles in many areas such as industrial manufacturing, disaster rescuing, unmanned vehicle, and science exploration. Because of the uncertain environment, changed resource, or dynamic system structure at runtime, traditional methods such as testing, model checking, and static analysis used in the development stage are not enough to ensure that the executions of robot control software satisfy specified properties. In this paper, we propose a runtime verification approach called Robots Monitor on Multilayer (RMoM) based on robot operating system (ROS) for monitoring whether the running of a robot swarm violates given temporal properties. To monitor robot system in a comprehensive manner over multiple layers, RMoM unifies resource, communication, robot, and swarm properties into a systemic, hierarchical monitoring framework. A discrete-time Metric Temporal Logic (MTL3) RMoM is proposed for specifying properties with timed and parameterized characteristics in robot swarms. Then, the corresponding three-valued semantics is defined for MTL3-RMoM to generate impartial and anticipatory monitors. Moreover, a hierarchical monitoring specification language high-level specification language (HSL)-RMoM and a series of monitor construction algorithms are proposed to automatically generate monitors for MTL3-RMoM properties on ROS platform. The experiments show that the method can automatically generate the monitors for detecting properties of robot swarms.

Yufan Jiang, Chi Hu, Tong Xiao, Chunliang Zhang, Jingbo Zhu. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

According to articulatory phonology, the gestural score is an invariant speech representation. Though the timing schemes, i.e., the onsets and offsets, of \nthe gestural activations may vary, the ensemble of these activations tends to remain unchanged, informing the speech content. "Gestural pattern vector" \n(GPV) has been proposed to encode the instantaneous gestural activations that exist across all tract variables at each time. Therefore, a gestural score with a particular timing scheme can be approximated using a GPV sequence. \n \nIn this work, we propose a pronunciation modeling method that uses a finite state machine (FSM) to represent the invariance of a gestural score. Given the "canonical" gestural score of a word with a known activation timing \nscheme, the plausible activation onsets and offsets are recursively generated and encoded as a weighted FSM. An empirical measure is used to prune out gestural activation timing schemes that deviate too much from the "canonical" gestural score. Speech recognition is achieved by matching the recovered \ngestural activations to the FSM-encoded gestural scores of different speech contents. In particular, the observation distribution of each GPV is modeled \nby an artificial neural network and Gaussian mixture tandem model. These models are used together with the FSM-based pronunciation models in a Bayesian framework. \n \nWe carry out pilot word classification experiments using synthesized data from one speaker. The proposed pronunciation modeling achieves over 90% accuracy for a vocabulary of 139 words with no training observations, outperforming direct use of the "canonical" gestural score.