Knowledge distillation (KD) has been applied to various tasks successfully, and mainstream methods typically boost the student model via spatial imitation losses. However, the consecutive downsamplings induced in the spatial domain of teacher model is a type of corruption, hindering the student from analyzing what specific information needs to be imitated, which results in accuracy degradation. To better understand the underlying pattern of corrupted feature maps, we shift our attention to the frequency domain. During frequency distillation, we encounter a new challenge: the low-frequency bands convey general but minimal context, while the high are more informative but also introduce noise. Not each pixel within the frequency bands contributes equally to the performance. To address the above problem: (1) We propose the Frequency Prompt plugged into the teacher model, absorbing the semantic frequency context during finetuning. (2) During the distillation period, a pixel-wise frequency mask is generated via Frequency Prompt, to localize those pixel of interests (PoIs) in various frequency bands. Additionally, we employ a position-aware relational frequency loss for dense prediction tasks, delivering a high-order spatial enhancement to the student model. We dub our Frequency Knowledge Distillation method as FreeKD, which determines the optimal localization and extent for the frequency distillation. Extensive experiments demonstrate that FreeKD not only outperforms spatial-based distillation methods consistently on dense prediction tasks (e.g., FreeKD brings 3.8 AP gains for RepPoints-R50 on COCO2017 and 4.55 mIoU gains for PSPNet-R18 on Cityscapes), but also conveys more robustness to the student. Notably, we also validate the generalization of our approach on large-scale vision models (e.g., DINO and SAM).
Membership Inference Attacks (MIA) aim to infer whether a target data record has been utilized for model training or not. Prior attempts have quantified the privacy risks of language models (LMs) via MIAs, but there is still no consensus on whether existing MIA algorithms can cause remarkable privacy leakage on practical Large Language Models (LLMs). Existing MIAs designed for LMs can be classified into two categories: reference-free and reference-based attacks. They are both based on the hypothesis that training records consistently strike a higher probability of being sampled. Nevertheless, this hypothesis heavily relies on the overfitting of target models, which will be mitigated by multiple regularization methods and the generalization of LLMs. The reference-based attack seems to achieve promising effectiveness in LLMs, which measures a more reliable membership signal by comparing the probability discrepancy between the target model and the reference model. However, the performance of reference-based attack is highly dependent on a reference dataset that closely resembles the training dataset, which is usually inaccessible in the practical scenario. Overall, existing MIAs are unable to effectively unveil privacy leakage over practical fine-tuned LLMs that are overfitting-free and private. We propose a Membership Inference Attack based on Self-calibrated Probabilistic Variation (SPV-MIA). Specifically, since memorization in LLMs is inevitable during the training process and occurs before overfitting, we introduce a more reliable membership signal, probabilistic variation, which is based on memorization rather than overfitting. Furthermore, we introduce a self-prompt approach, which constructs the dataset to fine-tune the reference model by prompting the target LLM itself. In this manner, the adversary can collect a dataset with a similar distribution from public APIs.
Integrated Sensing and Communication (ISAC) is a promising technology in 6G systems. The existing 3D Geometry-Based Stochastic Model (GBSM), as standardized for 5G systems, addresses solely communication channels and lacks consideration of the integration with sensing channel. Therefore, this letter extends 3D GBSM to support ISAC research, with a particular focus on capturing the sharing feature of both channels, including shared scatterers, clusters, paths, and similar propagation param-eters, which have been experimentally verified in the literature. The proposed approach can be summarized as follows: Firstly, an ISAC channel model is proposed, where shared and non-shared components are superimposed for both communication and sensing. Secondly, sensing channel is characterized as a cascade of TX-target, radar cross section, and target-RX, with the introduction of a novel parameter S for shared target extraction. Finally, an ISAC channel implementation framework is proposed, allowing flexible configuration of sharing feature and the joint generation of communication and sensing channels. The proposed ISAC channel model can be compatible with the 3GPP standards and offers promising support for ISAC technology evaluation.
This paper addresses an uplink localization problem in which the base station (BS) aims to locate a remote user with the aid of reconfigurable intelligent surface (RIS). This paper proposes a strategy in which the user transmits pilots over multiple time frames, and the BS adaptively adjusts the RIS reflection coefficients based on the observations already received so far in order to produce an accurate estimate of the user location at the end. This is a challenging active sensing problem for which finding an optimal solution involves a search through a complicated functional space whose dimension increases with the number of measurements. In this paper, we show that the long short-term memory (LSTM) network can be used to exploit the latent temporal correlation between measurements to automatically construct scalable information vectors (called hidden state) based on the measurements. Subsequently, the state vector can be mapped to the RIS configuration for the next time frame in a codebook-free fashion via a deep neural network (DNN). After all the measurements have been received, a final DNN can be used to map the LSTM cell state to the estimated user equipment (UE) position. Numerical result shows that the proposed active RIS design results in lower localization error as compared to existing active and nonactive methods. The proposed solution produces interpretable results and is generalizable to early stopping in the sequence of sensing stages.
Extremely large-scale multiple-input multiple-output (XL-MIMO) is a promising technology for the sixth-generation (6G) mobile communication networks. By significantly boosting the antenna number or size to at least an order of magnitude beyond current massive MIMO systems, XL-MIMO is expected to unprecedentedly enhance the spectral efficiency and spatial resolution for wireless communication. The evolution from massive MIMO to XL-MIMO is not simply an increase in the array size, but faces new design challenges, in terms of near-field channel modelling, performance analysis, channel estimation, and practical implementation. In this article, we give a comprehensive tutorial overview on near-field XL-MIMO communications, aiming to provide useful guidance for tackling the above challenges. First, the basic near-field modelling for XL-MIMO is established, by considering the new characteristics of non-uniform spherical wave (NUSW) and spatial non-stationarity. Next, based on the near-field modelling, the performance analysis of XL-MIMO is presented, including the near-field signal-to-noise ratio (SNR) scaling laws, beam focusing pattern, achievable rate, and degrees-of-freedom (DoF). Furthermore, various XL-MIMO design issues such as near-field beam codebook, beam training, channel estimation, and delay alignment modulation (DAM) transmission are elaborated. Finally, we point out promising directions to inspire future research on near-field XL-MIMO communications.
Zero-shot text-to-speech (TTS) synthesis aims to clone any unseen speaker's voice without adaptation parameters. By quantizing speech waveform into discrete acoustic tokens and modeling these tokens with the language model, recent language model-based TTS models show zero-shot speaker adaptation capabilities with only a 3-second acoustic prompt of an unseen speaker. However, they are limited by the length of the acoustic prompt, which makes it difficult to clone personal speaking style. In this paper, we propose a novel zero-shot TTS model with the multi-scale acoustic prompts based on a neural codec language model VALL-E. A speaker-aware text encoder is proposed to learn the personal speaking style at the phoneme-level from the style prompt consisting of multiple sentences. Following that, a VALL-E based acoustic decoder is utilized to model the timbre from the timbre prompt at the frame-level and generate speech. The experimental results show that our proposed method outperforms baselines in terms of naturalness and speaker similarity, and can achieve better performance by scaling out to a longer style prompt.
Membership Inference Attack (MIA) identifies whether a record exists in a machine learning model's training set by querying the model. MIAs on the classic classification models have been well-studied, and recent works have started to explore how to transplant MIA onto generative models. Our investigation indicates that existing MIAs designed for generative models mainly depend on the overfitting in target models. However, overfitting can be avoided by employing various regularization techniques, whereas existing MIAs demonstrate poor performance in practice. Unlike overfitting, memorization is essential for deep learning models to attain optimal performance, making it a more prevalent phenomenon. Memorization in generative models leads to an increasing trend in the probability distribution of generating records around the member record. Therefore, we propose a Probabilistic Fluctuation Assessing Membership Inference Attack (PFAMI), a black-box MIA that infers memberships by detecting these trends via analyzing the overall probabilistic fluctuations around given records. We conduct extensive experiments across multiple generative models and datasets, which demonstrate PFAMI can improve the attack success rate (ASR) by about 27.9% when compared with the best baseline.
Membership Inference Attack (MIA) identifies whether a record exists in a machine learning model's training set by querying the model. MIAs on the classic classification models have been well-studied, and recent works have started to explore how to transplant MIA onto generative models. Our investigation indicates that existing MIAs designed for generative models mainly depend on the overfitting in target models. However, overfitting can be avoided by employing various regularization techniques, whereas existing MIAs demonstrate poor performance in practice. Unlike overfitting, memorization is essential for deep learning models to attain optimal performance, making it a more prevalent phenomenon. Memorization in generative models leads to an increasing trend in the probability distribution of generating records around the member record. Therefore, we propose a Probabilistic Fluctuation Assessing Membership Inference Attack (PFAMI), a black-box MIA that infers memberships by detecting these trends via analyzing the overall probabilistic fluctuations around given records. We conduct extensive experiments across multiple generative models and datasets, which demonstrate PFAMI can improve the attack success rate (ASR) by about 27.9% when compared with the best baseline.
Video-based scene graph generation (VidSGG) is an approach that aims to represent video content in a dynamic graph by identifying visual entities and their relationships. Due to the inherently biased distribution and missing annotations in the training data, current VidSGG methods have been found to perform poorly on less-represented predicates. In this paper, we propose an explicit solution to address this under-explored issue by supplementing missing predicates that should be appear in the ground-truth annotations. Dubbed Trico, our method seeks to supplement the missing predicates by exploring three complementary spatio-temporal correlations. Guided by these correlations, the missing labels can be effectively supplemented thus achieving an unbiased predicate predictions. We validate the effectiveness of Trico on the most widely used VidSGG datasets, i.e., VidVRD and VidOR. Extensive experiments demonstrate the state-of-the-art performance achieved by Trico, particularly on those tail predicates.
We propose a novel approach for RGB-D salient instance segmentation using a dual-branch cross-modal feature calibration architecture called CalibNet. Our method simultaneously calibrates depth and RGB features in the kernel and mask branches to generate instance-aware kernels and mask features. CalibNet consists of three simple modules, a dynamic interactive kernel (DIK) and a weight-sharing fusion (WSF), which work together to generate effective instance-aware kernels and integrate cross-modal features. To improve the quality of depth features, we incorporate a depth similarity assessment (DSA) module prior to DIK and WSF. In addition, we further contribute a new DSIS dataset, which contains 1,940 images with elaborate instance-level annotations. Extensive experiments on three challenging benchmarks show that CalibNet yields a promising result, i.e., 58.0% AP with 320*480 input size on the COME15K-N test set, which significantly surpasses the alternative frameworks. Our code and dataset are available at: https://github.com/PJLallen/CalibNet.