We propose DA-MSIRNet, a lightweight yet innovative architecture for high-quality image inpainting that significantly enhances the standard U-Net through four key innovations: (1) Context Anchor Attention (CAA) for efficient global context modeling via adaptive region selection, (2) Sparse Self-Attention (SpA), inspired by Spa-former, for dynamic and precise local detail refinement by focusing on salient relationships, (3) Multi-Scale Inverted Residual (MSIR) modules for enhanced multi-scale feature fusion through optimized skip connections, and (4) Structural Similarity (SSIM) Loss for improved perceptual quality and fidelity. DA-MSIRNet effectively addresses critical limitations of existing methods, including GAN instability, U-Net’s restricted receptive field, and Transformer computational complexity. Comprehensive evaluations on Places2 and CelebA-HQ datasets demonstrate that DA-MSIRNet achieves state-of-the-art performance in both quantitative metrics (PSNR/SSIM/FID) and visual quality, while maintaining superior computational efficiency. The code for our DA-MSIRNet is publicly available on GitHub: https://github.com/nutcliu2507/DA-MSIRNet.

Wireless eavesdropping on phone conversations has become a major security and safety concern, especially with advancements toward 5G and beyond featuring higher frequencies and higher sensing resolution. As demonstrated recently, attackers can remotely detect even micron-scale acoustic vibrations emanating from a smartphone's earpiece via off-the-shelf millimeter-wave radar for audio information eavesdropping, all without the victim ever noticing. Here, we present a new architecture, MiSINFO, that not only thwarts such attacks but also enables the victim to counter-attack by spoofing of eavesdroppers with audio misinformation. With emerging attacks targeting the physical medium, i.e., acoustic signals, which cannot be protected by digital encryption and are the weakest segment of the communication chain, MiSINFO aims to systematically modify the eavesdroppers' fundamental sensing observations, concealing native signals while encoding alternate synthetic data. MiSINFO incorporates a low-profile, reconfigurable metasurface and double-inference principles to dynamically generate artificial audio-vibration signatures, injecting deceptive misinformation. We design, implement, and experimentally evaluate MiSINFO. Our results reveal that eavesdroppers detect none of the original words emitted by the speaker, while the injected misinformation is reconstructed with a low average word error rate of 2.29%. Our work represents the first such eavesdropping countermeasure which not only prevents attackers from accurately decoding the true signal but also uses a false signal to fool them into believing that they have succeeded. This approach transforms defensive measures from merely reactive to proactively deceptive, giving the defender an advantage and the capability to delude attackers into trusting false information.

Wireless eavesdropping on phone conversations has become a major security concern as attackers repurpose advanced wireless capabilities in 5G and beyond featuring higher frequencies and higher sensing resolution. Recent studies have demonstrated that attackers can exploit off-the-shelf millimeter-wave radars to covertly detect even micron-scale vibrations of smartphones caused by the earpiece during the phone conversation, eavesdropping on audio information without the victim ever noticing. In our IEEE S&P'25 paper, we present a new architecture that not only thwarts such attacks but also injects false signatures to fool eavesdroppers into believing they have succeeded. Here, we demonstrate the eavesdropping countermeasure technique that enables the user to hide his private acoustic signals and simultaneously inject an alternative signal via a low-profile, reconfigurable metasurface. We present a metasurface-based audio encoding method that generates artificial audio-vibration signatures to send deceptive audio information to eavesdroppers. We showcase experimental audio samples from both the attack and the proposed countermeasure, which transforms defensive strategies from merely reactive to proactively deceptive.

The next generation of wireless networks will utilize highly directional beams to overcome the path loss at high frequencies, requiring angle inference during link establishment. Furthermore, the integration of location-based services into the wireless infrastructure is rapidly increasing, bringing in a significant demand for an integrated fast localization scheme. In this work, we present a first-of-its-kind one-shot angular localization method that is carried out with a re-configurable architecture that unlocks ISAC functionality. Specifically, we use an electrically tunable metasurface with broadband response to generate wavefronts that are randomized across the angular space with diverse wideband amplitude and phase observations, corresponding to a collection of angle-unique one-shot beacons. Our results show down to 0.26° mean absolute error at 20 dB SNR, an order of magnitude improvement over the recently proposed one-shot solutions based on leaky-wave antennas (LWAs), in addition to having wider area coverage and less stringent bandwidth requirements.

Knowledge Distillation (KD) is a popular area of research for reducing the size of large models while still maintaining good performance.The outputs of larger teacher models are used to guide the training of smaller student models.Given the repetitive nature of acoustic events, we propose to leverage this information to regulate the KD training for Audio Tagging.This novel KD method, Intra-Utterance Similarity Preserving KD (IUSP), shows promising results for the audio tagging task.It is motivated by the previously published KD method: Similarity Preserving KD (SP).However, instead of preserving the pairwise similarities between inputs within a mini-batch, our method preserves the pairwise similarities between the frames of a single input utterance.Our proposed KD method, IUSP, shows consistent improvements over SP across student models of different sizes on the DCASE 2019 Task 5 dataset for audio tagging.There is a 27.1% to 122.4% percent increase in improvement of micro AUPRC over the baseline relative to SPs improvement of over the baseline.

Hybrid deep neural network hidden Markov models (DNN-HMM) have achieved impressive results on large vocabulary continuous speech recognition (LVCSR) tasks. However, the recent approaches using DNN-HMM models are not explored much for text recognition. Inspired by the current work in automatic speech recognition (ASR) and machine translation, we present an open vocabulary sub-word text recognition system. The sub-word lexicon and sub-word language model (LM) helps in overcoming the challenge of recognizing out of vocabulary (OOV) words, and a time delay neural network (TDNN) and convolution neural network (CNN) based DNN-HMM optical model (OM) efficiently models the sequence dependency in the line image. We present results on 12 datasets with training data varying from 6k lines to 600k lines. The system is built for 8 languages, i.e., English, French, Arabic, Chinese, Farsi, Tamil, Russian, and Korean. We report competitive results on several commonly used handwritten and printed text datasets.

This research examines an assessment system that combines a mobile device with NFC (Near Field Communication) technology. Teachers attach or embed NFC tags into answer choices and students answer each question by bringing their mobile phone close to the answer that they think is the correct one. A four-week experiment shows that this assessment method takes more time than a traditional one. Also, the grades differ from those using traditional assessment. Analysis of the experiment results and subsequent tester survey and interviews show that this assessment method can help students incorporate what they have learned in the classroom into real life, as well as reflects how well students comprehend what is taught to them in a classroom setting.

In multihop MANETs, nodes have to cooperate to forward each other's packets through the networks. Every node, including source and intermediate nodes, has a fair opportunity to transmit a packet Thus, the hot spot may suffer traffic congestion. The packet loss rate and the transmission delay are increased, but the throughput is decreased Due to the contention for the shared channel, the throughput of each single node is limited not only by the channel capacity, but also by the transmissions in its neighborhood (intraflow/interflow contention problem). Furthermore, the network throughput still restrict by the channel capacity. Some other related works using multiple channels simultaneously to transmit packets without interfering each other increases the throughput But if the traffic load is heavy, these schemes may induce more serious packet loss on the contrary because they do not consider the congestion problem especially in multihop MANETs. In this paper, we propose a new multichannel MAC protocol using multichannel transmission, and develop a hop- by-hop congestion control scheme, which is named "Efficient Flow Control with Multichannels" (EFCM). The EFCM scheme modifies the IEEE 802.11 RTS/CTS frame format to reserve bandwidth in different channels. It also solves the hidden terminal problem in the multichannel environment The intermediate nodes have a higher priority than the source (or leaf) nodes to contend for the right of transmission to solve the intraflow contention. In order to solve the interflow contention, congestion control is taken in every node to construct a flow table, which restricts the number of packets of every flow passing by congested nodes. According to the simulation results, the whole network throughputs of EFCM are improved significantly.