arXiv subjects

Jialong Zhang

Publications and source records attributed to Jialong Zhang.

3 recordsLinked to original sources

arXiv2026

On the Fundamental Limits of Single-snapshot Compressive Ultrasound Imaging Using Random Aberrative Masks

Compressive sensing emerges as a paradigm shift to realize real-time volumetric ultrasound imaging using a single element transducer equipped with an aberrative mask. Such a coded aperture encodes each scatterer in the FoV as a specific echo signal, compressing each volumetric into a time sequence that are reconstructed into image via computational methods. Compressive imaging (CI) can greatly reduce the data rate and simplify the electronics. Despite this potential, the practical performance of single-snapshot CI remains unclear, especially how it is influenced by mask design, imaging task complexity, and reconstruction strategy. For this, we separate the information budget provided by the mask from the algorithm-dependent extraction of the encoded information. First, the spatial impulse responses of random masks with different pixel sizes and time-delay ranges were calibrated, and the entropy-based effective rank of their similarity matrix was used to quantify the available encoding capacity. Masks with pixel size of approx. half wavelength and a time-delay range of two carrier periods provided higher encoding capacities, reaching up to 1.1% of the total sampled voxels. Algorithm-dependent information extraction was further evaluated for both ULM-motivated particle localization and B-mode imaging. For particle localization, L1-norm regularized least-squares method reconstructed particles corresponding to approx. 10% of the available encoding capacity, outperforming the matched filter. A transition from successful to failed reconstruction was observed when the number of particles exceeded this value, revealing the upper limit of the present CI systems using random masks. For B-mode imaging, LSQR achieved the highest SSIM among the evaluated methods, of up to 0.12, although the encoding capacity of the current random masks remained insufficient for high-fidelity reconstruction.

physics.app-ph↗

arXiv2020

Confidential Inference via Ternary Model Partitioning

Today's cloud vendors are competing to provide various offerings to simplify and accelerate AI service deployment. However, cloud users always have concerns about the confidentiality of their runtime data, which are supposed to be processed on third-party's compute infrastructures. Information disclosure of user-supplied data may jeopardize users' privacy and breach increasingly stringent data protection regulations. In this paper, we systematically investigate the life cycles of inference inputs in deep learning image classification pipelines and understand how the information could be leaked. Based on the discovered insights, we develop a Ternary Model Partitioning mechanism and bring trusted execution environments to mitigate the identified information leakages. Our research prototype consists of two co-operative components: (1) Model Assessment Framework, a local model evaluation and partitioning tool that assists cloud users in deployment preparation; (2) Infenclave, an enclave-based model serving system for online confidential inference in the cloud. We have conducted comprehensive security and performance evaluation on three representative ImageNet-level deep learning models with different network depths and architectural complexity. Our results demonstrate the feasibility of launching confidential inference services in the cloud with maximized confidentiality guarantees and low performance costs.

cs.CR↗

arXiv2018

Reaching Data Confidentiality and Model Accountability on the CalTrain

Distributed collaborative learning (DCL) paradigms enable building joint machine learning models from distrusting multi-party participants. Data confidentiality is guaranteed by retaining private training data on each participant's local infrastructure. However, this approach to achieving data confidentiality makes today's DCL designs fundamentally vulnerable to data poisoning and backdoor attacks. It also limits DCL's model accountability, which is key to backtracking the responsible "bad" training data instances/contributors. In this paper, we introduce CALTRAIN, a Trusted Execution Environment (TEE) based centralized multi-party collaborative learning system that simultaneously achieves data confidentiality and model accountability. CALTRAIN enforces isolated computation on centrally aggregated training data to guarantee data confidentiality. To support building accountable learning models, we securely maintain the links between training instances and their corresponding contributors. Our evaluation shows that the models generated from CALTRAIN can achieve the same prediction accuracy when compared to the models trained in non-protected environments. We also demonstrate that when malicious training participants tend to implant backdoors during model training, CALTRAIN can accurately and precisely discover the poisoned and mislabeled training data that lead to the runtime mispredictions.

cs.CR↗