Modern machine learning models toward various tasks with omic data analysis give rise to threats of privacy leakage patients involved in those datasets. Here, we proposed a secure and privacy-preserving method (PPML-Omics) by designing decentralized differential private federated algorithm. We applied PPML-Omics analyze from three sequencing technologies addressed the concern major under representative deep models. examined breaches depth through attack experiments demonstrated that could...

Auditing algorithms' privacy typically involves simulating a game-based protocol that guesses which of two adjacent datasets was the original input. Traditional approaches require thousands such simulations, leading to significant computational overhead. Recent methods propose single-run auditing target algorithm address this, substantially reducing cost. However, these methods' general applicability and tightness in producing empirical guarantees remain uncertain. This work studies problems...

Privacy and Byzantine resilience are two indispensable requirements for a federated learning (FL) system. Although there have been extensive studies on privacy security in their own track, solutions that consider both remain sparse. This is due to difficulties reconciling privacy-preserving Byzantine-resilient algorithms. In this work, we propose solution such two-fold issue. We use our version of differentially private stochastic gradient descent (DP-SGD) algorithm preserve then apply note...

Abstract Modern machine learning models towards various tasks with omic data analysis give rise to threats of privacy leakage patients involved in those datasets. Despite the advances different technologies, existing methods tend introduce too much computational cost (e.g. cryptographic methods) or noise differential privacy), which hampers either model usefulness accuracy protecting biological data. Here, we proposed a secure and privacy-preserving method (PPML-Omics) by designing...

We study the bias introduced in Differentially-Private Stochastic Gradient Descent (DP-SGD) with clipped or normalized per-sample gradient. As one of most popular but artificial operations to ensure bounded sensitivity, gradient clipping enables composite privacy analysis many iterative optimization methods without additional assumptions on either learning models input data. Despite its wide applicability, also presents theoretical challenges systematically instructing improvement utility....

In this paper, we propose a sensor with high torsion sensitivity. This has the ability to distinguish direction. A long-period fiber grating (LPFG) unique geometric structure is used configure sensor. The proposed LPFG prepared by CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> laser polishing method and etching method. First, periodically polish two opposite planes which are parallel axis direction of get sheets structure. Then,...

In this letter, a highly sensitive bending sensor based on an embedded multimode D-shaped long period fiber grating (EMD-LPFG) is proposed. The novel applied to carry out vector measurement. proposed LPFG fabricated by polishing the prepared structure which formed periodically splicing between single mode (SMF) and (MMF). Since cross section of MMF D-shaped, we named it EMD-LPFG. Due asymmetric modulation refractive index CO 2 laser, has ability distinguish directions, MMFs provide higher...

In this paper, a novel liquid level sensor with high sensitivity is proposed. The proposed configured by multimode fiber-No-core fiber-multimode fiber (MMF-NCF-MMF) hetero core structure (MNM-HCS). composed of two MMF structures spliced on both ends NCF, in which NCF used as sensing unit, and the at excites higher cladding modes to improve external environment. Interference occurs when light transmitted from NCF. Experimental results show that compact length 12.2 mm has 1034 pm/mm water (RI...

We proposed and experimentally demonstrated a vector bending sensor based on core-offset long period fiber grating (CO-LPFG). The CO-LPFG is configured by CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> laser polishing three mutually orthogonal planes of single mode (SMF) to different depths then heating it the electrical arc-discharge fusion splicer. asymmetrical cladding structure leads an uneven density distribution in molten...

Abstract In this paper, we propose an arc-remodified long period fiber grating (AR-LPFG) and introduce the ratios of shift-distance two resonant dips (RSRDs) as a novel distinguishing index to monitor bending absolute condition which refers detect curvature directions under stationary status. The AR-LPFG is configured by arc-remodifying LPFG that prepared from CO 2 laser. refractive modulated again arc-remodification method results in new dips. Two orthogonal are distinguished comparing...

In this paper, we proposed a novel long-period fiber grating strain sensor based on local ellipse-core (LEC-LPFG). The LEC-LPFG was fabricated by laser polishing and heating methods, which could efficiently transform single-mode (SMF) into periodic fiber. effects associated with the elliptical core shape were shown to significantly influence sensitivity of LEC-LPFG. Further, it demonstrated that has high −20.2 pm/ <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML"...

In this paper, a novel liquid level sensor with ultra-high sensitivity is proposed. The proposed configured by slice-shaped composite long period fiber grating (SSC-LPFG). SSC-LPFG prepared polishing two opposite sides of multimode–single-mode–multimode structure using CO2 laser. method improves the to external environment. Based on simulation calculation, length 3 mm designed. experimental transmission spectrum agrees well result. results show that reaches 7080 pm/mm in range 0–1400 μm...

We study the problem of guaranteeing Differential Privacy (DP) in hyper-parameter tuning, a crucial process machine learning involving selection best run from several. Unlike many private algorithms, including prevalent DP-SGD, privacy implications tuning remain insufficiently understood. Recent works propose generic solution for process, yet fundamental question still persists: is current bound this tight? This paper contributes both positive and negative answers to question. Initially, we...

In this letter, a novel strain sensor based on four-sided modulated long period fiber grating (FM-LPFG) is introduced and demonstrated. Due to the asymmetric refractive index modulation of CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> laser, cladding radius reduced core deviation effect strengthened. addition, by simulating equivalent during tension, sensitive area FM-LPFG proposed. The measurement results indicate that sensitivity...

A torsion sensor based on intermittent-spiral long period fiber grating (IS-LPFG) is proposed the interaction mechanism of elliptical birefringence (IMEB). The IS-LPFG fabricated by periodically polishing four mutually orthogonal surfaces a pre-twisted single-mode (SMF) CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> laser. caused distribution and individual spiral structure determines performance sensing. sensitivity reaches to 0.15...

Privacy and Byzantine resilience are two indispensable requirements for a federated learning (FL) system. Although there have been extensive studies on privacy security in their own track, solutions that consider both remain sparse. This is due to difficulties reconciling privacy-preserving Byzantine-resilient algorithms. In this work, we propose solution such two-fold issue. We use our version of differentially private stochastic gradient descent (DP-SGD) algorithm preserve then apply note...

In this paper, we study the (sparse) Generalized Eigenvalue Problem (GEP), which arises in a number of modern statistical learning models, such as principal component analysis (PCA), canonical correlation (CCA), Fisher's discriminant (FDA) and sliced inverse regression (SIR). We provide first on GEP differential privacy (DP) model under both deterministic stochastic settings. low dimensional case, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Differential privacy (DP) has seen immense applications in learning on tabular, image, and sequential data where instance-level is concerned. In graphs, contrastingly, works node-level are highly sparse. Challenges arise as existing DP protocols hardly apply to the message-passing mechanism Graph Neural Networks (GNNs). this study, we propose a solution that specifically addresses issue of privacy. Our protocol consists two main components: 1) sampling routine called HeterPoisson, which...

In this paper, a novel strain sensor based on four-sided modulated long period fiber grating (FM-LPFG) is introduced and demonstrated. Due to the asymmetric refractive index modulation of CO 2 laser, cladding radius reduced core deviation effect strengthened. addition, by simulating equivalent during tension, sensitive area FM-LPFG proposed. The measurement results indicate that sensitivity 127.5 pm/με (0~125 με) 17.7 (125~575 με). proposed can provide high capability precision measurement.

Abstract A novel refractive-index (RI) sensor based on a D-shaped modulated single-mode-graded-index multimode-single-mode fiber (DM-SGIMS) structure is presented and investigated in detail. CO 2 laser was used to process the SGIMS structure, which produced strong evanescent field. The structural parameters were determined via simulations, experimental results consistent with those of theoretical analysis. transmission spectra RI sensitivity fabricated sensors different lengths compared...