arXiv subjects

Yinliang Diao

Publications and source records attributed to Yinliang Diao.

6 recordsLinked to original sources

arXiv2025

Breaking the Memory Wall: Exact Analytical Differentiation via Tiled Operator-Space Evolution

Selective State Space Models (SSMs) achieve linear-time inference, yet their gradient-based sensitivity analysis remains bottlenecked by O(L) memory scaling during backpropagation. This memory constraint precludes genomic-scale modeling (L > 10^5) on consumer-grade hardware. We introduce Phase Gradient Flow (PGF), a framework that computes exact analytical derivatives by operating directly in the state-space manifold, bypassing the need to materialize the intermediate computational graph. By reframing SSM dynamics as Tiled Operator-Space Evolution (TOSE), our method delivers O(1) memory complexity relative to sequence length, yielding a 94% reduction in peak VRAM and a 23x increase in throughput compared to standard Autograd. Unlike parallel prefix scans that exhibit numerical divergence in stiff ODE regimes, PGF ensures stability through invariant error scaling, maintaining near-machine precision across extreme sequences. We demonstrate the utility of PGF on an impulse-response benchmark with 128,000-step sequences - a scale where conventional Autograd encounters prohibitive memory overhead, often leading to out-of-memory (OOM) failures in multi-layered models. Our work enables chromosome-scale sensitivity analysis on a single GPU, bridging the gap between theoretical infinite-context models and practical hardware limitations.

cs.LG↗

arXiv2025

Electromagnetic Field Exposure Assessment and Mitigation Strategies for Wireless Power Transfer Systems: A Review and Future Perspectives

Wireless power transfer (WPT) technologies are increasingly being applied in fields ranging from consumer electronics and electric vehicles to space-based energy systems and medical implants. While WPT offers contactless power delivery, it introduces electromagnetic field (EMF) emissions, necessitating careful assessment to address safety and public health concerns. Exposure guidelines developed by ICNIRP and IEEE define frequency-dependent limits based on internal quantities, such as electric field strength and specific absorption rate, intended to prevent tissue nerve stimulation < 100 kHz and heating > 100 kHz, respectively. Complementing these guidelines, assessment standards including the International Electrotechnical Commission (IEC)/IEEE 63184 and IEC Technical Report 63377, provide practical procedures for evaluating the EMF exposure in WPT systems. This review offers a comparative overview of major WPT modalities, with a focus on recent developments in computational dosimetry and standardized assessment techniques for the complex, non-uniform fields typical of WPT environments. It also discusses electromagnetic interference with medical devices and exposure scenarios involving partial body proximity and various postures. A notable observation across modalities is the considerable variability, often spanning an order of magnitude, in the allowable transfer power, depending on the field distribution and assessment approach. Remaining challenges include the lack of harmonized guidance for intermediate frequencies and localized exposure, underscoring the importance of further coordination in international standardization efforts. Addressing these issues is essential for the safe and widespread deployment of WPT technologies.

physics.app-ph↗

arXiv2025

Roadmap towards Personalized Approaches and Safety Considerations in Non-Ionizing Radiation: From Dosimetry to Therapeutic and Diagnostic Applications

This roadmap provides a comprehensive and forward-looking perspective on the individualized application and safety of non-ionizing radiation (NIR) dosimetry in diagnostic and therapeutic medicine. Covering a wide range of frequencies, i.e., from low-frequency to terahertz, this document provides an overview of the current state of the art and anticipates future research needs in selected key topics of NIR-based medical applications. It also emphasizes the importance of personalized dosimetry, rigorous safety evaluation, and interdisciplinary collaboration to ensure safe and effective integration of NIR technologies in modern therapy and diagnosis.

physics.med-ph↗

arXiv2020

Assessment of absorbed power density and temperature rise for nonplanar body model under electromagnetic exposure above 6 GHz

The averaged absorbed power density (APD) and temperature rise in body models with nonplanar surfaces were computed for electromagnetic exposure above 6 GHz. Different calculation schemes for the averaged APD were investigated. Additionally, a novel compensation method for correcting the heat convection rate on the air/skin interface in voxel human models was proposed and validated. The compensation method can be easily incorporated into bioheat calculations and does not require information regarding the normal direction of the boundary voxels, in contrast to a previously proposed method. The APD and temperature rise were evaluated using models of a two-dimensional cylinder and a three-dimensional partial forearm. The heating factor, which was defined as the ratio of the temperature rise to the APD, was calculated using different APD averaging schemes. Our computational results revealed different frequency and curvature dependences. For body models with curvature radii of >30 mm and at frequencies of >20 GHz, the differences in the heating factors among the APD schemes were small.

physics.comp-ph↗

arXiv2020

Effect of skin-to-skin contact on stimulation threshold and dosimetry

The human dosimetry for electromagnetic field exposure is an essential task to develop exposure guidelines/standards for human safety as well as product safety assessment. At frequencies from a few hundred Hz to 10 MHz, the adverse effect to be protected is the stimulation of the peripheral nervous system. The in situ electric field in the skin is used as a surrogate of nerve activation. In the low-frequency dosimetry, a high but inaccurate in situ electric field has been reported at positions where a skin-to-skin contact exists, whose relation to the stimulation is controversial. One of the reasons for high electric fields may be attributable to the current resolution of anatomical models. In this study, we first evaluate the stimulation threshold at postures of skin-to-skin contact experimentally for different hand/finger positions to represent skin touching/non-touching scenarios. We confirm that the skin-to-skin contact does not lower the threshold current of magnetic stimulation devices needed to induce pain. Second, a new method is proposed for hand modeling to configure different finger positions using static hand models with similar postures to the experiments. We compute the in situ electric field at skin-to-skin contact for the different hand posture scenarios that indicate an excessive raise of the electric field in skin-to-skin regions that is not justified by the experiments. The comparison suggests that a high in situ electric field in the skin would be caused by poor modeling of the skin layers, which is not enough to represent in a resolution of the order of a millimeter. This skin-to-skin contact should not be considered to set the restriction in the international exposure guidelines/standards as well as product safety assessment.

physics.med-ph↗

arXiv2020

Learning-based estimation of dielectric properties and tissue density in head models for personalized radio-frequency dosimetry

Radio-frequency dosimetry is an important process in human safety and for compliance of related products. Recently, computational human models generated from medical images have often been used for such assessment, especially to consider the inter-variability of subjects. However, the common procedure to develop personalized models is time consuming because it involves excessive segmentation of several components that represent different biological tissues, which limits the inter-variability assessment of radiation safety based on personalized dosimetry. Deep learning methods have been shown to be a powerful approach for pattern recognition and signal analysis. Convolutional neural networks with deep architecture are proven robust for feature extraction and image mapping in several biomedical applications. In this study, we develop a learning-based approach for fast and accurate estimation of the dielectric properties and density of tissues directly from magnetic resonance images in a single shot. The smooth distribution of the dielectric properties in head models, which is realized using a process without tissue segmentation, improves the smoothness of the specific absorption rate (SAR) distribution compared with that in the commonly used procedure. The estimated SAR distributions, as well as that averaged over 10-g of tissue in a cubic shape, are found to be highly consistent with those computed using the conventional methods that employ segmentation.

cs.LG↗