Search

In practical scenarios, federated learning frequently necessitates training personalized models for each client using heterogeneous data. This paper proposes a backbone self-distillation approach to facilitate personalized federated learning. In this approach, each client trains its local model and only sends the backbone weights to the server. These weights are then aggregated to create a global backbone, which is returned to each client for updating. However, the client's local backbone lacks personalization because of the common representation. To solve this problem, each client further performs backbone self-distillation by using the global backbone as a teacher and transferring knowledge to update the local backbone. This process involves learning two components: the shared backbone for common representation and the private head for local personalization, which enables effective global knowledge transfer. Extensive experiments and comparisons with 12 state-of-the-art approaches demonstrate the effectiveness of our approach., Comment: Pubished in ACM MMAsia 2023

Federated learning provides a privacy-preserving manner to collaboratively train models on data distributed over multiple local clients via the coordination of a global server. In this paper, we focus on label distribution skew in federated learning, where due to the different user behavior of the client, label distributions between different clients are significantly different. When faced with such cases, most existing methods will lead to a suboptimal optimization due to the inadequate utilization of label distribution information in clients. Inspired by this, we propose a label-masking distillation approach termed FedLMD to facilitate federated learning via perceiving the various label distributions of each client. We classify the labels into majority and minority labels based on the number of examples per class during training. The client model learns the knowledge of majority labels from local data. The process of distillation masks out the predictions of majority labels from the global model, so that it can focus more on preserving the minority label knowledge of the client. A series of experiments show that the proposed approach can achieve state-of-the-art performance in various cases. Moreover, considering the limited resources of the clients, we propose a variant FedLMD-Tf that does not require an additional teacher, which outperforms previous lightweight approaches without increasing computational costs. Our code is available at https://github.com/wnma3mz/FedLMD., Comment: Accepted by ACM MM 2023

Deep learning models can achieve high inference accuracy by extracting rich knowledge from massive well-annotated data, but may pose the risk of data privacy leakage in practical deployment. In this paper, we present an effective teacher-student learning approach to train privacy-preserving deep learning models via differentially private data-free distillation. The main idea is generating synthetic data to learn a student that can mimic the ability of a teacher well-trained on private data. In the approach, a generator is first pretrained in a data-free manner by incorporating the teacher as a fixed discriminator. With the generator, massive synthetic data can be generated for model training without exposing data privacy. Then, the synthetic data is fed into the teacher to generate private labels. Towards this end, we propose a label differential privacy algorithm termed selective randomized response to protect the label information. Finally, a student is trained on the synthetic data with the supervision of private labels. In this way, both data privacy and label privacy are well protected in a unified framework, leading to privacy-preserving models. Extensive experiments and analysis clearly demonstrate the effectiveness of our approach., Comment: Published by IEEE MMSP 2022

While deep models have proved successful in learning rich knowledge from massive well-annotated data, they may pose a privacy leakage risk in practical deployment. It is necessary to find an effective trade-off between high utility and strong privacy. In this work, we propose a discriminative-generative distillation approach to learn privacy-preserving deep models. Our key idea is taking models as bridge to distill knowledge from private data and then transfer it to learn a student network via two streams. First, discriminative stream trains a baseline classifier on private data and an ensemble of teachers on multiple disjoint private subsets, respectively. Then, generative stream takes the classifier as a fixed discriminator and trains a generator in a data-free manner. After that, the generator is used to generate massive synthetic data which are further applied to train a variational autoencoder (VAE). Among these synthetic data, a few of them are fed into the teacher ensemble to query labels via differentially private aggregation, while most of them are embedded to the trained VAE for reconstructing synthetic data. Finally, a semi-supervised student learning is performed to simultaneously handle two tasks: knowledge transfer from the teachers with distillation on few privately labeled synthetic data, and knowledge enhancement with tangent-normal adversarial regularization on many triples of reconstructed synthetic data. In this way, our approach can control query cost over private data and mitigate accuracy degradation in a unified manner, leading to a privacy-preserving student model. Extensive experiments and analysis clearly show the effectiveness of the proposed approach., Comment: This paper is accepted by IEEE Transactions on Image Processing (TIP)

While the success of deep learning relies on large amounts of training datasets, data is often limited in privacy-sensitive domains. To address this challenge, generative model learning with differential privacy has emerged as a solution to train private generative models for desensitized data generation. However, the quality of the images generated by existing methods is limited due to the complexity of modeling data distribution. We build on the success of diffusion models and introduce DP-SAD, which trains a private diffusion model by a stochastic adversarial distillation method. Specifically, we first train a diffusion model as a teacher and then train a student by distillation, in which we achieve differential privacy by adding noise to the gradients from other models to the student. For better generation quality, we introduce a discriminator to distinguish whether an image is from the teacher or the student, which forms the adversarial training. Extensive experiments and analysis clearly demonstrate the effectiveness of our proposed method., Comment: accepted by ECCV 2024

We propose a method based on the Hilbert-Huang transform (HHT) to recover the high-energy waveform of low-frequency quasi-periodic oscillations (LFQPOs). Based on the method, we successfully obtain the modulation of the phase-folded light curve above 170 keV using the QPO phase reconstructed at lower energies in MAXI J1535-571 with Insight-HXMT observations. A comprehensive simulation study is conducted to demonstrate that such modulation indeed originates from the QPO. Thus the highest energies turn out to significantly exceed the upper limit of ~100 keV for QPOs reported previously using the Fourier method, marking the first opportunity to study QPO properties above 100 keV in this source. Detailed analyses of these high-energy QPO profiles reveal different QPO properties between the 30-100 keV and 100-200 keV energy ranges: the phase lag remains relatively stable, and the amplitude slightly increases below ~100 keV, whereas above this threshold, soft phase lags and a decrease in amplitude are observed. Given the reports of a hard tail detection in broad spectroscopy, we propose that the newly discovered QPO properties above 100 keV are dominated by the hard tail component, possibly stemming from a relativistic jet. Our findings also indicate a strong correlation between the QPOs originating from the jet and corona, supporting the scenario of jet-corona coupling precssion. We emphasize that our proposed HHT-based method can serve as an efficient manner in expanding the high energy band for studying QPOs, thereby enhancing our understanding of their origin., Comment: 21 pages, 15 figures, accepted for publication in ApJL

We present a detailed analysis of the X-ray luminosity (Lx) dependence of the cyclotron absorption line energy (Ecyc) for the X-ray binary pulsar 1A 0535+262 during its 2020 giant outburst based on pulse-to-pulse analysis. By applying this technique to high cadence observations of Insight-HXMT, we reveal the most comprehensive Ecyc-Lx correlation across a broad luminosity range of ~(0.03-1.3)*10^38 erg/s. Apart from the positive and negative correlations between cyclotron line energy and luminosity at Lx~(1-3)*10^37 erg/s and ~(7-13)*10^37 erg/s, which are expected from the typical subcritical and supercritical accretion regimes, respectively, a plateau in the correlation is also detected at ~(3-7)*10^37 erg/s^-1. Moreover, at the lowest luminosity level (Lx<10^37 erg/s), the positive Ecyc-Lx correlation seems to be broken, and the pulse profile also occurs a significant transition. These discoveries provide the first complete view on the correlation between luminosity and the centriod energy of the cyclotron line, and therefore are relevant for understanding how accretion onto magnetized neutron stars depends on luminosity., Comment: Accepted for publication in Monthly Notices of the Royal Astronomical Society Main Journal

Training state-of-the-art (SOTA) deep models often requires extensive data, resulting in substantial training and storage costs. To address these challenges, dataset condensation has been developed to learn a small synthetic set that preserves essential information from the original large-scale dataset. Nowadays, optimization-oriented methods have been the primary method in the field of dataset condensation for achieving SOTA results. However, the bi-level optimization process hinders the practical application of such methods to realistic and larger datasets. To enhance condensation efficiency, previous works proposed Distribution-Matching (DM) as an alternative, which significantly reduces the condensation cost. Nonetheless, current DM-based methods still yield less comparable results to SOTA optimization-oriented methods. In this paper, we argue that existing DM-based methods overlook the higher-order alignment of the distributions, which may lead to sub-optimal matching results. Inspired by this, we present a novel DM-based method named M3D for dataset condensation by Minimizing the Maximum Mean Discrepancy between feature representations of the synthetic and real images. By embedding their distributions in a reproducing kernel Hilbert space, we align all orders of moments of the distributions of real and synthetic images, resulting in a more generalized condensed set. Notably, our method even surpasses the SOTA optimization-oriented method IDC on the high-resolution ImageNet dataset. Extensive analysis is conducted to verify the effectiveness of the proposed method. Source codes are available at https://github.com/Hansong-Zhang/M3D., Comment: This work has been accepted by AAAI-24

We conduct a spectral and timing analysis of GX 339-4 and EXO 1846-031 with the aim of studying the evolution of Type-C QPOs with spectral parameters. The high cadence data from Insight-HXMT and NICER allow us to track them. Type-C QPOs appear at the end of low-hard state and/or hard-intermediate state. The results reveal that the QPO frequency is closely related to the inner disk radius and mass accretion rate in the two sources. Such a correlation is nicely consistent with the dynamic frequency model., Comment: 14 pages, 15 figures. v2: refereed version

While generative models have proved successful in many domains, they may pose a privacy leakage risk in practical deployment. To address this issue, differentially private generative model learning has emerged as a solution to train private generative models for different downstream tasks. However, existing private generative modeling approaches face significant challenges in generating high-dimensional data due to the inherent complexity involved in modeling such data. In this work, we present a new private generative modeling approach where samples are generated via Hamiltonian dynamics with gradients of the private dataset estimated by a well-trained network. In the approach, we achieve differential privacy by perturbing the projection vectors in the estimation of gradients with sliced score matching. In addition, we enhance the reconstruction ability of the model by incorporating a residual enhancement module during the score matching. For sampling, we perform Hamiltonian dynamics with gradients estimated by the well-trained network, allowing the sampled data close to the private dataset's manifold step by step. In this way, our model is able to generate data with a resolution of 256x256. Extensive experiments and analysis clearly demonstrate the effectiveness and rationality of the proposed approach., Comment: accepted by ACM MM 2024 Oral

While massive valuable deep models trained on large-scale data have been released to facilitate the artificial intelligence community, they may encounter attacks in deployment which leads to privacy leakage of training data. In this work, we propose a learning approach termed differentially private data-free distillation (DPDFD) for model conversion that can convert a pretrained model (teacher) into its privacy-preserving counterpart (student) via an intermediate generator without access to training data. The learning collaborates three parties in a unified way. First, massive synthetic data are generated with the generator. Then, they are fed into the teacher and student to compute differentially private gradients by normalizing the gradients and adding noise before performing descent. Finally, the student is updated with these differentially private gradients and the generator is updated by taking the student as a fixed discriminator in an alternate manner. In addition to a privacy-preserving student, the generator can generate synthetic data in a differentially private way for other downstream tasks. We theoretically prove that our approach can guarantee differential privacy and well convergence. Extensive experiments clearly demonstrate that our approach significantly outperform other differentially private generative approaches., Comment: Published at IJCAI 2023

We studied the millihertz quasi-periodic oscillation (mHz QPO) in the 2020 outburst of the Be/X-ray binary 1A 0535+262 using Insight-HXMT data over a broad energy band. The mHz QPO is detected in the 27-120 keV energy band. The QPO centroid frequency is correlated with the source flux, and evolves in the 35-95 mHz range during the outburst. The QPO is most significant in the 50-65 keV band, with a significance of ~ 8 sigma, but is hardly detectable (<2 sigma) in the lowest (1-27 keV) and highest (>120 keV) energy bands. Notably, the detection of mHz QPO above 80 keV is the highest energy at which mHz QPOs have been detected so far. The fractional rms of the mHz QPO first increases and then decreases with energy, reaching the maximum amplitude at 50-65 keV. In addition, at the peak of the outburst, the mHz QPO shows a double-peak structure, with the difference between the two peaks being constant at ~0.02 Hz, twice the spin frequency of the neutron star in this system. We discuss different scenarios explaining the generation of the mHz QPO, including the beat frequency model, the Keplerian frequency model, the model of two jets in opposite directions, and the precession of the neutron star, but find that none of them can explain the origin of the QPO well. We conclude that the variability of non-thermal radiation may account for the mHz QPO, but further theoretical studies are needed to reveal the physical mechanism., Comment: 13 pages, 7 figures. Accepted for publication in MNRAS

While the abuse of deepfake technology has caused serious concerns recently, how to detect deepfake videos is still a challenge due to the high photo-realistic synthesis of each frame. Existing image-level approaches often focus on single frame and ignore the spatiotemporal cues hidden in deepfake videos, resulting in poor generalization and robustness. The key of a video-level detector is to fully exploit the spatiotemporal inconsistency distributed in local facial regions across different frames in deepfake videos. Inspired by that, this paper proposes a simple yet effective patch-level approach to facilitate deepfake video detection via spatiotemporal dropout transformer. The approach reorganizes each input video into bag of patches that is then fed into a vision transformer to achieve robust representation. Specifically, a spatiotemporal dropout operation is proposed to fully explore patch-level spatiotemporal cues and serve as effective data augmentation to further enhance model's robustness and generalization ability. The operation is flexible and can be easily plugged into existing vision transformers. Extensive experiments demonstrate the effectiveness of our approach against 25 state-of-the-arts with impressive robustness, generalizability, and representation ability.

We study the spectral evolution of the black hole candidate EXO 1846$-$031 during its 2019 outburst, in the 1--150 keV band,with the {\it {Hard X-ray Modulation Telescope}}. The continuum spectrum is well modelled with an absorbed disk-blackbody plus cutoff power-law, in the hard, intermediate and soft states. In addition, we detect an $\approx$6.6 keV Fe emission line in the hard intermediate state. Throughout the soft intermediate and soft states, the fitted inner disk radius remains almost constant; we suggest that it has settled at the innermost stable circular orbit (ISCO). However, in the hard and hard intermediate states, the apparent inner radius was unphysically small (smaller than ISCO), even after accounting for the Compton scattering of some of the disk photons by the corona in the fit. We argue that this is the result of a high hardening factor, $f_{\rm col}\approx2.0-2.7$, in the early phases of outburst evolution, well above the canonical value of 1.7 suitable to a steady disk. We suggest that the inner disk radius was close to ISCO already in the low/hard state. Furthermore, we propose that this high value of hardening factor in the relatively hard state is probably caused by the additional illuminating of the coronal irradiation onto the disk. Additionally, we estimate the spin parameter with the continuum-fitting method, over a range of plausible black hole masses and distances. We compare our results with the spin measured with the reflection-fitting method and find that the inconsistency of the two results is partly caused by the different choices of $f_{\rm col}$., Comment: 21 pages, 6 figures

The origin(s) and mechanism(s) of fast radio bursts (FRBs), which are short radio pulses from cosmological distances, have remained a major puzzle since their discovery. We report a strong Quasi-Periodic Oscillation(QPO) of 40 Hz in the X-ray burst from the magnetar SGR J1935+2154 and associated with FRB 200428, significantly detected with the Hard X-ray Modulation Telescope (Insight-HXMT) and also hinted by the Konus-Wind data. QPOs from magnetar bursts have only been rarely detected; our 3.4 sigma (p-value is 2.9e-4) detection of the QPO reported here reveals the strongest QPO signal observed from magnetars (except in some very rare giant flares), making this X-ray burst unique among magnetar bursts. The two X-ray spikes coinciding with the two FRB pulses are also among the peaks of the QPO. Our results suggest that at least some FRBs are related to strong oscillation processes of neutron stars. We also show that we may overestimate the significance of the QPO signal and underestimate the errors of QPO parameters if QPO exists only in a fraction of the time series of a X-ray burst which we use to calculate the Leahy-normalized periodogram., Comment: 17 pages, 13 figures, accepted for publication by ApJ. arXiv admin note: text overlap with arXiv:1212.1011 by other authors. text overlap with arXiv:1212.1011 by other authors

Why should we trust the detections of deep neural networks for manipulated faces? Understanding the reasons is important for users in improving the fairness, reliability, privacy and trust of the detection models. In this work, we propose an interpretable face manipulation detection approach to achieve the trustworthy and accurate inference. The approach could make the face manipulation detection process transparent by embedding the feature whitening module. This module aims to whiten the internal working mechanism of deep networks through feature decorrelation and feature constraint. The experimental results demonstrate that our proposed approach can strike a balance between the detection accuracy and the model interpretability.

The need for dietary data management is growing with public awareness of food intakes. As a result, there are increasing deployments of smart canteens where dietary data is collected through either Radio Frequency Identification (RFID) or Computer Vision(CV)-based solutions. As human labor is involved in both cases, manpower allocation is critical to data quality. Where manpower requirements are underestimated, data quality is compromised. This paper has studied the relation between the quality of dietary data and the manpower invested, using numerical simulations based on real data collected from multiple smart canteens. We found that in both RFID and CV-based systems, the long-term cost of dietary data acquisition is dominated by manpower. Our study provides a comprehensive understanding of the cost composition for dietary data acquisition and useful insights toward future cost effective systems., Comment: Will be presented on the Computing Conference 2020, London, Jul. 2020