Ni Ding

Ni Ding, Parastoo Sadeghi, Rodney A. Kennedy

We study the adaptive modulation (AM) problem in a network-coded two-way relay channel (NC-TWRC), where each of the two users controls its own bit rate in the $m$-ary quadrature amplitude modulation ($m$-QAM) to minimize the transmission error rate and enhance the spectral efficiency. We show that there exists a strategic complementarity, one user tends to transmit while the other decides to do so in order to enhance the overall spectral efficiency, which is beyond the scope of the conventional single-agent AM scheduling method. We propose a two-player game model parameterized by the signal-to-noise ratios (SNRs) of two user-to-user channels and prove that it is a supermodular game where there always exist the extremal pure strategy Nash equilibria (PSNEs), the largest and smallest PSNEs. We show by simulation results that the extremal PSNEs incur a similar bit error rate (BER) as the conventional single-agent AM scheme, but significantly improve the spectral efficiency in the NC-TWRC system. The study also reveals the Pareto order of the extremal PSNEs: The largest and smallest PSNEs are Pareto worst and best PSNEs, respectively. Finally, we derive the sufficient conditions for the extremal PSNEs to be symmetric and monotonic in channel SNRs. We also discuss how to utilize the symmetry and monotonicity to relieve the complexity in the PSNE learning process.

Siyuan Wen, Jiahao Zeng, Ningning Ding

Diffusion unlearning is essential for mitigating the generation of harmful or copyrighted content in text-to-image models. Current diffusion unlearning techniques determine the model update direction by either using alternatives of the target concept as an anchor or using empty prompts. The anchor-based method relies on manually and semantically-chosen anchors that risk biased unlearning, while the anchor-free method inherently suffers from unrobust unlearning due to unconstrained latent updates. In this work, we theoretically formalize such unstable diffusion unlearning issues under the manifold hypothesis and prove that lacking a manifold-proximal anchor inevitably induces significant normal-space drift that degrades unlearning performance. To achieve stable unlearning, we propose \mysysn, a two-stage framework that automatically synthesizes manifold-proximal anchors. However, direct geometric manifold optimization is computationally intractable. To address this challenge, \mysys introduces a novel cross-attention consistency loss which serves as a highly efficient surrogate of manifold proximity. Experimental results demonstrate that \mysys effectively achieves robust and unbiased unlearning across various state-of-the-art baselines, significantly improving targeted concept removal (by up to 31.04\% in CLIP score) and non-target utility (by up to 4.18\% in CLIP score). Moreover, \mysys can also be easily integrated into existing diffusion unlearning methods to enhance their unlearning performance (by 6.30\% for concept removal and 6.65\% for utility on average).

Yuru Wang, Lejun Cheng, Yuxin Zuo et al.

We introduce NatureBench, a cross-discipline benchmark of 90 tasks distilled from peer-reviewed Nature-family publications, designed to evaluate whether AI coding agents can move beyond reproduction toward discovery on real scientific problems. NatureBench is built on NatureGym, an automated pipeline that constructs a standardized, per-task containerized environment from a source paper, addressing the environment-fragmentation problem that has limited the credibility of prior agent-on-research benchmarks. Evaluating ten frontier agent configurations under a strict web-search-disabled protocol, we find that the strongest model surpasses SOTA on only 17.8% of tasks under the g>0.1 criterion. Analysis of method pathways reveals that agents succeed primarily through methodological translation, converting scientific tasks into familiar supervised prediction problems, rather than through genuine scientific invention. Failures are dominated by wrong method choice and insufficient compute budget, not by task misunderstanding. We release the benchmark, the NatureGym pipeline, and a public leaderboard with maintainer-side reproduction. Code: https://github.com/FrontisAI/NatureBench

Yuxin Zuo, Zikai Xiao, Li Sheng et al.

A world model predicts environment dynamics based on current observations and actions, serving as a core cognitive mechanism for reasoning and planning. In this work, we investigate how world modeling based on language models can further push the boundaries of general agents. (i) We first focus on building foundation models for agentic environment simulation. We introduce Qwen-AgentWorld-35B-A3B and Qwen-AgentWorld-397B-A17B, the first language world models capable of simulating agentic environments covering 7 domains via long chain-of-thought reasoning. Leveraging more than 10M environment interaction trajectories of 7 domains in real-world environments, we develop Qwen-AgentWorld through a three-stage training pipeline: CPT injects general-purpose world modeling capabilities from the state transition dynamics and augmented professional corpora, SFT activates next-state-prediction reasoning, and RL sharpens simulation fidelity through a tailored framework with hybrid rubric-and-rule rewards. To evaluate language world models, we present AgentWorldBench, a comprehensive benchmark constructed from real-world interactions of 5 frontier models on 9 established benchmarks. Empirical results demonstrate that Qwen-AgentWorld significantly outperforms existing frontier models. (ii) Beyond foundation models, we further investigate two complementary paradigms through which world modeling enhances general agents. First, as a decoupled environment simulator, Qwen-AgentWorld supports scalable and controllable simulation of thousands of real-world environments for agentic RL, yielding gains that surpass real-environment training alone. Second, as a unified agent foundation model, world-model training acts as a highly effective warm-up that improves downstream performance across 7 agentic benchmarks. Code: https://github.com/QwenLM/Qwen-AgentWorld

Shaoshan Liu, Xiugong Qin, Xuan Wu et al.

The scalability of humanoid robots will depend not only on models and hardware, but also on whether physical experience can accumulate across robots, tasks, organizations, and time. Drawing on the authors' work in developing ISO/WD 26264-1, Humanoid robot datasets -- Part 1: General requirements, within ISO/TC 299/WG 16, this article argues that data standards are becoming foundational infrastructure for Physical AI. We develop three insights. First, humanoid robot data is embodied interaction data, not a collection of isolated digital samples; a useful dataset must preserve the relationship among robot body, action, task, scene, execution trace, and outcome. Second, its value depends on physical coherence: multimodal streams are reusable only when timing, coordinate frames, calibration, kinematics, units, and synchronization assumptions remain inspectable. Third, the main bottleneck is not only data scarcity, but non-cumulative data caused by high collection costs, data silos, and inconsistent evaluation. We argue that humanoid robot data standards address these bottlenecks by making embodied experience interpretable, shareable, traceable, and reusable. A general standard should provide horizontal infrastructure for lifecycle management, metadata, provenance, quality, versioning, and traceability, while capability-specific parts should define domain grammar for manipulation, locomotion, human-robot interaction, cognition, and future humanoid capabilities. As AI moves from screens into bodies, data standards must evolve from organizing digital information to structuring physical interaction.

Ni Ding, Lei He, Shengbo Eben Li et al.

End-to-end autonomous driving has emerged as a dominant paradigm, yet its highly entangled black-box models pose significant challenges in terms of interpretability and safety assurance. To improve model transparency and training flexibility, this paper proposes a hierarchical and decoupled post-training framework tailored for pretrained neural networks. By reconstructing intermediate feature maps from ground-truth labels, surrogate supervisory signals are introduced at transitional layers to enable independent training of specific components, thereby avoiding the complexity and coupling of conventional end-to-end backpropagation and providing interpretable insights into networks' internal mechanisms. To the best of our knowledge, this is the first method to formalize feature-level reverse computation as well-posed optimization problems, which we rigorously reformulate as systems of linear equations or least squares problems. This establishes a novel and efficient training paradigm that extends gradient backpropagation to feature backpropagation. Extensive experiments on multiple standard image classification benchmarks demonstrate that the proposed method achieves superior generalization performance and computational efficiency compared to traditional training approaches, validating its effectiveness and potential.

Ni Ding, Miao Qiao, Jiaxing Xu et al.

This paper proposes $α$-GAN, a generative adversarial network using Rényi measures. The value function is formulated, by Rényi cross entropy, as an expected certainty measure incurred by the discriminator's soft decision as to where the sample is from, true population or the generator. The discriminator tries to maximize the Rényi certainty about sample source, while the generator wants to reduce it by injecting fake samples. This forms a min-max problem with the solution parameterized by the Rényi order $α$. This $α$-GAN reduces to vanilla GAN at $α= 1$, where the value function is exactly the binary cross entropy. The optimization of $α$-GAN is over probability (vector) space. It is shown that the gradient is exponentially enlarged when Rényi order is in the range $α\in (0,1)$. This makes convergence faster, which is verified by experimental results. A discussion shows that choosing $α\in (0,1)$ may be able to solve some common problems, e.g., vanishing gradient. A following observation reveals that this range has not been fully explored in the existing Rényi version GANs.

Farhad Farokhi, Ni Ding

We propose an operational measure of information leakage in a non-stochastic setting to formalize privacy against a brute-force guessing adversary. We use uncertain variables, non-probabilistic counterparts of random variables, to construct a guessing framework in which an adversary is interested in determining private information based on uncertain reports. We consider brute-force trial-and-error guessing in which an adversary can potentially check all the possibilities of the private information that are compatible with the available outputs to find the actual private realization. The ratio of the worst-case number of guesses for the adversary in the presence of the output and in the absence of it captures the reduction in the adversary's guessing complexity and is thus used as a measure of private information leakage. We investigate the relationship between the newly-developed measure of information leakage with the existing non-stochastic maximin information and stochastic maximal leakage that are shown arise in one-shot guessing.

Ni Ding, Farhad Farokhi

We consider a non-stochastic privacy-preserving problem in which an adversary aims to infer sensitive information $S$ from publicly accessible data $X$ without using statistics. We consider the problem of generating and releasing a quantization $\hat{X}$ of $X$ to minimize the privacy leakage of $S$ to $\hat{X}$ while maintaining a certain level of utility (or, inversely, the quantization loss). The variables $S$ and $S$ are treated as bounded and non-probabilistic, but are otherwise general. We consider two existing non-stochastic privacy measures, namely the maximum uncertainty reduction $L_0(S \rightarrow \hat{X})$ and the refined information $I_*(S; \hat{X})$ (also called the maximin information) of $S$. For each privacy measure, we propose a corresponding agglomerative clustering algorithm that converges to a locally optimal quantization solution $\hat{X}$ by iteratively merging elements in the alphabet of $X$. To instantiate the solution to this problem, we consider two specific utility measures, the worst-case resolution of $X$ by observing $\hat{X}$ and the maximal distortion of the released data $\hat{X}$. We show that the value of the maximin information $I_*(S; \hat{X})$ can be determined by dividing the confusability graph into connected subgraphs. Hence, $I_*(S; \hat{X})$ can be reduced by merging nodes connecting subgraphs. The relation to the probabilistic information-theoretic privacy is also studied by noting that the G{á}cs-K{ö}rner common information is the stochastic version of $I_*$ and indicates the attainability of statistical indistinguishability.

Ni Ding, Parastoo Sadeghi, Rodney A. Kennedy

This paper considers a cross-layer adaptive modulation system that is modeled as a Markov decision process (MDP). We study how to utilize the monotonicity of the optimal transmission policy to relieve the computational complexity of dynamic programming (DP). In this system, a scheduler controls the bit rate of the m-quadrature amplitude modulation (m-QAM) in order to minimize the long-term losses incurred by the queue overflow in the data link layer and the transmission power consumption in the physical layer. The work is done in two steps. Firstly, we observe the L-natural-convexity and submodularity of DP to prove that the optimal policy is always nondecreasing in queue occupancy/state and derive the sufficient condition for it to be nondecreasing in both queue and channel states. We also show that, due to the L-natural-convexity of DP, the variation of the optimal policy in queue state is restricted by a bounded marginal effect: The increment of the optimal policy between adjacent queue states is no greater than one. Secondly, we use the monotonicity results to present two low complexity algorithms: monotonic policy iteration (MPI) based on L-natural-convexity and discrete simultaneous perturbation stochastic approximation (DSPSA). We run experiments to show that the time complexity of MPI based on L-natural-convexity is much lower than that of DP and the conventional MPI that is based on submodularity and DSPSA is able to adaptively track the optimal policy when the system parameters change.

Ni Ding, Parastoo Sadeghi, Rodney A. Kennedy

This paper considers the discrete convexity of a cross-layer on-off transmission control problem in wireless communications. In this system, a scheduler decides whether or not to transmit in order to optimize the long-term quality of service (QoS) incurred by the queueing effects in the data link layer and the transmission power consumption in the physical (PHY) layer simultaneously. Using a Markov decision process (MDP) formulation, we show that the optimal policy can be determined by solving a minimization problem over a set of queue thresholds if the dynamic programming (DP) is submodular. We prove that this minimization problem is discrete convex. In order to search the minimizer, we consider two discrete stochastic approximation (DSA) algorithms: discrete simultaneous perturbation stochastic approximation (DSPSA) and L-natural-convex stochastic approximation (L-natural-convex SA). Through numerical studies, we show that the two DSA algorithms converge significantly faster than the existing continuous simultaneous perturbation stochastic approximation (CSPSA) algorithm in multi-user systems. Finally, we compare the convergence results and complexity of two DSA and CSPSA algorithms where we show that DSPSA achieves the best trade-off between complexity and accuracy in multi-user systems.

Ni Ding, Parastoo Sadeghi, Rodney A. Kennedy

This paper considers a transmission control problem in network-coded two-way relay channels (NC-TWRC), where the relay buffers random symbol arrivals from two users, and the channels are assumed to be fading. The problem is modeled by a discounted infinite horizon Markov decision process (MDP). The objective is to find a transmission control policy that minimizes the symbol delay, buffer overflow and transmission power consumption and error rate simultaneously and in the long run. By using the concepts of submodularity, multimodularity and L-natural convexity, we study the structure of the optimal policy searched by dynamic programming (DP) algorithm. We show that the optimal transmission policy is nondecreasing in queue occupancies or/and channel states under certain conditions such as the chosen values of parameters in the MDP model, channel modeling method, modulation scheme and the preservation of stochastic dominance in the transitions of system states. The results derived in this paper can be used to relieve the high complexity of DP and facilitate real-time control.