Emotion recognition has broad applications in fields such as healthcare and human-computer interaction. As effective biomarkers reflecting emotional states, physiological signals have attracted significant attention. However, cross-subject emotion recognition based on physiological signals, especially multimodal signals, faces three major challenges: sample heterogeneity, modality heterogeneity, and subject heterogeneity. To address these issues, we propose a Multilevel Augmented Dual Contrastive Learning Network (MADCL-Net). MADCL-Net mitigates sample-level differences through hierarchical data augmentation and reduces modality and subject discrepancies by introducing dual contrastive learning at both the modality and subject levels. Two novel contrastive loss functions are designed for this purpose. An attention module is further integrated to enhance the complementarity among different modalities. Extensive experiments on the DEAP and DREAMER datasets demonstrate the superiority of our approach. Specifically, MADCL-Net outperforms the second-best state-of-the-art model by 1.53% in the arousal dimension on the DEAP dataset and by 2.65% in the valence dimension on the DREAMER dataset, while achieving the second-best performance in the other dimension on both datasets.

The rapid progress of navigation, manipulation, and vision models has made mobile manipulators capable in many specialized tasks. However, the open-world mobile manipulation (OWMM) task remains a challenge due to the need for generalization to open-ended instructions and environments, as well as the systematic complexity to integrate high-level decision making with low-level robot control based on both global scene understanding and current agent state. To address this complexity, we propose a novel multi-modal agent architecture that maintains multi-view scene frames and agent states for decision-making and controls the robot by function calling. A second challenge is the hallucination from domain shift. To enhance the agent performance, we further introduce an agentic data synthesis pipeline for the OWMM task to adapt the VLM model to our task domain with instruction fine-tuning. We highlight our fine-tuned OWMM-VLM as the first dedicated foundation model for mobile manipulators with global scene understanding, robot state tracking, and multi-modal action generation in a unified model. Through experiments, we demonstrate that our model achieves SOTA performance compared to other foundation models including GPT-4o and strong zero-shot generalization in real world. The project page is at https://hhyhrhy.github.io/owmm-agent-project.

To boost the encoder stealing attack under the perturbation-based defense that hinders the attack performance, we propose a boosting encoder stealing attack with perturbation recovery named BESA. It aims to overcome perturbation-based defenses. The core of BESA consists of two modules: perturbation detection and perturbation recovery, which can be combined with canonical encoder stealing attacks. The perturbation detection module utilizes the feature vectors obtained from the target encoder to infer the defense mechanism employed by the service provider. Once the defense mechanism is detected, the perturbation recovery module leverages the well-designed generative model to restore a clean feature vector from the perturbed one. Through extensive evaluations based on various datasets, we demonstrate that BESA significantly enhances the surrogate encoder accuracy of existing encoder stealing attacks by up to 24.63% when facing state-of-the-art defenses and combinations of multiple defenses.

Phase noise is known as a main bottleneck in full-duplex (FD) communications, and limits the self-interference (SI) cancellation capability. There exist some works to mitigate the negative impact of phase noise, but usually relying on a perfect self-interference (SI) channel response information, which is difficult to be estimated accurately because of phase noise behavior. To improve the SI cancellation capability and suppress the impact of phase noise, we propose a multiple-downconversion FD transceiver design, which includes a new FD transceiver architecture and an SI cancellation algorithm. The proposed FD transceiver architecture deploys multiple receive chains to downconvert the signal split from the receive antenna. Particularly, the oscillator signal of each receive chain is originated from the transmit oscillator with a unique delay, such that the phase noises contained in the multipath SI components can be partially compensated. The SI cancellation algorithm includes a phase noise coefficients extraction step and an SI channel estimation step. As the extracted phase noise coefficients are used in the SI channel estimation, the estimation accuracy is improved. Simulation results show that, the limit of cancellation capability of the proposed FD transceiver with the proposed SI channel estimator is improved by 10 dB compared to that of the proposed FD transceiver with least square (LS) channel estimator.

Federated learning (FL) has achieved state-of-the-art performance in distributed learning tasks with privacy requirements. However, it has been discovered that FL is vulnerable to adversarial attacks. The typical gradient inversion attacks primarily focus on attempting to obtain the client’s private input in a white-box manner, where the adversary is assumed to be either the client or the server. However, if both the clients and the server are honest and fully trusted, is the FL secure? In this paper, we propose a novel method called External Gradient Inversion Attack (EGIA) in the grey-box settings. Specifically, we concentrate on the point that public-shared gradients in FL are always transmitted through the intermediary nodes, which has been widely ignored. On this basis, we demonstrate that an external adversary can reconstruct the private input using gradients even if both the clients and the server are honest and fully trusted. We also provide a comprehensive theoretical analysis of the black-box attack scenario in which the adversary has only the gradients. We perform extensive experiments on multiple real-world datasets to test the effectiveness of EGIA. The outcomes of our experiments validate that the EGIA method is highly effective.

Knowing model parameters has been regarded as a vital factor for recovering sensitive information from the gradients in federated learning. But is it safe to use federated learning when the model parameters are unavailable for adversaries, i.e., external adversaries’ In this paper, we answer this question by proposing a novel gradient inversion attack. Speciffically, we observe a widely ignored fact in federated learning that the participants’ gradient data are usually transmitted via the intermediary node. Based on this fact, we show that an external adversary is able to recover the private input from the gradients, even if it does not have the model parameters. Through extensive experiments based on several real-world datasets, we demonstrate that our proposed new attack can recover the input with pixelwise accuracy and feasible efficiency.