-
Small Objects Matters in Weakly-supervised Semantic Segmentation
Authors:
Cheolhyun Mun,
Sanghuk Lee,
Youngjung Uh,
Junsuk Choe,
Hyeran Byun
Abstract:
Weakly-supervised semantic segmentation (WSSS) performs pixel-wise classification given only image-level labels for training. Despite the difficulty of this task, the research community has achieved promising results over the last five years. Still, current WSSS literature misses the detailed sense of how well the methods perform on different sizes of objects. Thus we propose a novel evaluation me…
▽ More
Weakly-supervised semantic segmentation (WSSS) performs pixel-wise classification given only image-level labels for training. Despite the difficulty of this task, the research community has achieved promising results over the last five years. Still, current WSSS literature misses the detailed sense of how well the methods perform on different sizes of objects. Thus we propose a novel evaluation metric to provide a comprehensive assessment across different object sizes and collect a size-balanced evaluation set to complement PASCAL VOC. With these two gadgets, we reveal that the existing WSSS methods struggle in capturing small objects. Furthermore, we propose a size-balanced cross-entropy loss coupled with a proper training strategy. It generally improves existing WSSS methods as validated upon ten baselines on three different datasets.
△ Less
Submitted 25 September, 2023;
originally announced September 2023.
-
Influence of M/A substitution on material properties of intermetallic compounds MSn$_2$ (M = Fe, Co; A = Li, Na): A first-principles study
Authors:
Chol-Jun Yu,
Un-Song Hwang,
Yong-Chol Pak,
Kyonga Rim,
Chol Ryu,
Chon-Ryong Mun,
Un-Gi Jong
Abstract:
Iron and cobalt distannides \ce{MSn2} (M = Fe, Co) are regarded as a promising conversion-type anode material for lithium- and sodium-ion batteries, but their properties are not well understood. In this work, we report a first-principles study of alkali metal (A = Li, Na) substitutional effect on the structural, mechanical, lattice vibrational, electronic and defect properties of these distannides…
▽ More
Iron and cobalt distannides \ce{MSn2} (M = Fe, Co) are regarded as a promising conversion-type anode material for lithium- and sodium-ion batteries, but their properties are not well understood. In this work, we report a first-principles study of alkali metal (A = Li, Na) substitutional effect on the structural, mechanical, lattice vibrational, electronic and defect properties of these distannides. Special attention is paid to systematic comparison between \ce{FeSn2} and \ce{CoSn2}. Our calculations reveal that M/A substitution induces a lattice expansion and decrease of elastic constants, which is more announced with Na substitution than Li, and moreover changes the elastic property of \ce{FeSn2} from ductile to brittle whereas preserves the ductility of \ce{CoSn2}. An imaginary phonon frequency mode appears only for \ce{FeSn2} and \ce{FeNaSn2}, and M/A substitution provokes a definite gap between high and low frequency regions. We perform a careful analysis of electronic density of states, band structures and Fermi surface, providing an insight into difference of electronic structures between \ce{FeSn2} and \ce{CoSn2}. With further calculation of defect formation energies and alkali ion diffusion barriers, we believe this work can be useful to design conversion-type anode materials for alkali-ion batteries.
△ Less
Submitted 23 July, 2020;
originally announced July 2020.
-
PrivFT: Private and Fast Text Classification with Homomorphic Encryption
Authors:
Ahmad Al Badawi,
Luong Hoang,
Chan Fook Mun,
Kim Laine,
Khin Mi Mi Aung
Abstract:
The need for privacy-preserving analytics is higher than ever due to the severity of privacy risks and to comply with new privacy regulations leading to an amplified interest in privacy-preserving techniques that try to balance between privacy and utility. In this work, we present an efficient method for Text Classification while preserving the privacy of the content using Fully Homomorphic Encryp…
▽ More
The need for privacy-preserving analytics is higher than ever due to the severity of privacy risks and to comply with new privacy regulations leading to an amplified interest in privacy-preserving techniques that try to balance between privacy and utility. In this work, we present an efficient method for Text Classification while preserving the privacy of the content using Fully Homomorphic Encryption (FHE). Our system (named \textbf{Priv}ate \textbf{F}ast \textbf{T}ext (PrivFT)) performs two tasks: 1) making inference of encrypted user inputs using a plaintext model and 2) training an effective model using an encrypted dataset. For inference, we train a supervised model and outline a system for homomorphic inference on encrypted user inputs with zero loss to prediction accuracy. In the second part, we show how to train a model using fully encrypted data to generate an encrypted model. We provide a GPU implementation of the Cheon-Kim-Kim-Song (CKKS) FHE scheme and compare it with existing CPU implementations to achieve 1 to 2 orders of magnitude speedup at various parameter settings. We implement PrivFT in GPUs to achieve a run time per inference of less than 0.66 seconds. Training on a relatively large encrypted dataset is more computationally intensive requiring 5.04 days.
△ Less
Submitted 18 November, 2019; v1 submitted 18 August, 2019;
originally announced August 2019.
-
CaRENets: Compact and Resource-Efficient CNN for Homomorphic Inference on Encrypted Medical Images
Authors:
** Chao,
Ahmad Al Badawi,
Balagopal Unnikrishnan,
Jie Lin,
Chan Fook Mun,
James M. Brown,
J. Peter Campbell,
Michael Chiang,
Jayashree Kalpathy-Cramer,
Vijay Ramaseshan Chandrasekhar,
Pavitra Krishnaswamy,
Khin Mi Mi Aung
Abstract:
Convolutional neural networks (CNNs) have enabled significant performance leaps in medical image classification tasks. However, translating neural network models for clinical applications remains challenging due to data privacy issues. Fully Homomorphic Encryption (FHE) has the potential to address this challenge as it enables the use of CNNs on encrypted images. However, current HE technology pos…
▽ More
Convolutional neural networks (CNNs) have enabled significant performance leaps in medical image classification tasks. However, translating neural network models for clinical applications remains challenging due to data privacy issues. Fully Homomorphic Encryption (FHE) has the potential to address this challenge as it enables the use of CNNs on encrypted images. However, current HE technology poses immense computational and memory overheads, particularly for high-resolution images such as those seen in the clinical context. We present CaRENets: Compact and Resource-Efficient CNNs for high performance and resource-efficient inference on high-resolution encrypted images in practical applications. At the core, CaRENets comprises a new FHE compact packing scheme that is tightly integrated with CNN functions. CaRENets offers dual advantages of memory efficiency (due to compact packing of images and CNN activations) and inference speed (due to the reduction in the number of ciphertexts created and the associated mathematical operations) over standard interleaved packing schemes. We apply CaRENets to perform homomorphic abnormality detection with 80-bit security level in two clinical conditions - Retinopathy of Prematurity (ROP) and Diabetic Retinopathy (DR). The ROP dataset comprises 96 x 96 grayscale images, while the DR dataset comprises 256 x 256 RGB images. We demonstrate over 45x improvement in memory efficiency and 4-5x speedup in inference over the interleaved packing schemes. As our approach enables memory-efficient low-latency HE inference without imposing additional communication burden, it has implications for practical and secure deep learning inference in clinical imaging.
△ Less
Submitted 28 January, 2019;
originally announced January 2019.
-
Towards the AlexNet Moment for Homomorphic Encryption: HCNN, theFirst Homomorphic CNN on Encrypted Data with GPUs
Authors:
Ahmad Al Badawi,
** Chao,
Jie Lin,
Chan Fook Mun,
Jun Jie Sim,
Benjamin Hong Meng Tan,
Xiao Nan,
Khin Mi Mi Aung,
Vijay Ramaseshan Chandrasekhar
Abstract:
Deep Learning as a Service (DLaaS) stands as a promising solution for cloud-based inference applications. In this setting, the cloud has a pre-learned model whereas the user has samples on which she wants to run the model. The biggest concern with DLaaS is user privacy if the input samples are sensitive data. We provide here an efficient privacy-preserving system by employing high-end technologies…
▽ More
Deep Learning as a Service (DLaaS) stands as a promising solution for cloud-based inference applications. In this setting, the cloud has a pre-learned model whereas the user has samples on which she wants to run the model. The biggest concern with DLaaS is user privacy if the input samples are sensitive data. We provide here an efficient privacy-preserving system by employing high-end technologies such as Fully Homomorphic Encryption (FHE), Convolutional Neural Networks (CNNs) and Graphics Processing Units (GPUs). FHE, with its widely-known feature of computing on encrypted data, empowers a wide range of privacy-concerned applications. This comes at high cost as it requires enormous computing power. In this paper, we show how to accelerate the performance of running CNNs on encrypted data with GPUs. We evaluated two CNNs to classify homomorphically the MNIST and CIFAR-10 datasets. Our solution achieved a sufficient security level (> 80 bit) and reasonable classification accuracy (99%) and (77.55%) for MNIST and CIFAR-10, respectively. In terms of latency, we could classify an image in 5.16 seconds and 304.43 seconds for MNIST and CIFAR-10, respectively. Our system can also classify a batch of images (> 8,000) without extra overhead.
△ Less
Submitted 18 August, 2020; v1 submitted 2 November, 2018;
originally announced November 2018.
-
Codebook-Based SDMA for Coexistence with Fixed Wireless Service
Authors:
Han-Shin Jo,
Cheol Mun,
Jong-Gwan Yook
Abstract:
A portion of frequency band for International Mobile Telecommunications (IMT)-Advanced is currently allocated to Fixed Wireless Service (FWS) such as Fixed Service (FS), Fixed Satellite Service (FSS), or Fixed Wireless Access (FWA), which requires frequency sharing between both the systems. SDMA, due to its high throughput nature, is candidate for IMT-Advanced. This paper proposes a systematic des…
▽ More
A portion of frequency band for International Mobile Telecommunications (IMT)-Advanced is currently allocated to Fixed Wireless Service (FWS) such as Fixed Service (FS), Fixed Satellite Service (FSS), or Fixed Wireless Access (FWA), which requires frequency sharing between both the systems. SDMA, due to its high throughput nature, is candidate for IMT-Advanced. This paper proposes a systematic design of a precoder codebook for SDMA sharing spectrum with existing FWS. Based on an estimated direction angle of a victim FWS system, an interfering transmitter adaptively constructs a codebook forming a transmit null in the direction angle while satisfying orthogonal beamforming constraint. We derive not only asymptotic throughput scaling laws, but also an upperbound on throughput loss to analyze performance loss of the proposed SDMA relative to the popular SDMA called per-user unitary rate control (PU2RC). Furthermore, we develop a method of evaluating protection distance in order to analyze the spectrum sharing performance of the proposed approach. The simulation results of protection distance confirm that the proposed SDMA efficiently shares spectrum with FWS systems by reducing protection distance to more than 66%. Although our proposed SDMA always has lower throughput compared to PU2RC in non-coexistence scenario, it offers an intriguing opportunity to reuse spectrum already allocated to FWS.
△ Less
Submitted 20 October, 2010; v1 submitted 16 June, 2010;
originally announced June 2010.
-
Interference Mitigation Using Uplink Power Control for Two-Tier Femtocell Networks
Authors:
Han-Shin Jo,
Cheol Mun,
June Moon,
Jong-Gwan Yook
Abstract:
This paper proposes two interference mitigation strategies that adjust the maximum transmit power of femtocell users to suppress the cross-tier interference at a macrocell base station (BS). The open-loop and the closed-loop control suppress the cross-tier interference less than a fixed threshold and an adaptive threshold based on the noise and interference (NI) level at the macrocell BS, respec…
▽ More
This paper proposes two interference mitigation strategies that adjust the maximum transmit power of femtocell users to suppress the cross-tier interference at a macrocell base station (BS). The open-loop and the closed-loop control suppress the cross-tier interference less than a fixed threshold and an adaptive threshold based on the noise and interference (NI) level at the macrocell BS, respectively. Simulation results show that both schemes effectively compensate the uplink throughput degradation of the macrocell BS due to the cross-tier interference and that the closed-loop control provides better femtocell throughput than the open-loop control at a minimal cost of macrocell throughput.
△ Less
Submitted 12 October, 2009;
originally announced October 2009.
-
Self-optimized Coverage Coordination in Femtocell Networks
Authors:
Han-Shin Jo,
Cheol Mun,
June Moon,
Jong-Gwan Yook
Abstract:
This paper proposes a self-optimized coverage coordination scheme for two-tier femtocell networks, in which a femtocell base station adjusts the transmit power based on the statistics of the signal and the interference power that is measured at a femtocell downlink. Furthermore, an analytic expression is derived for the coverage leakage probability that a femtocell coverage area leaks into an outd…
▽ More
This paper proposes a self-optimized coverage coordination scheme for two-tier femtocell networks, in which a femtocell base station adjusts the transmit power based on the statistics of the signal and the interference power that is measured at a femtocell downlink. Furthermore, an analytic expression is derived for the coverage leakage probability that a femtocell coverage area leaks into an outdoor macrocell. The coverage analysis is verified by simulation, which shows that the proposed scheme provides sufficient indoor femtocell coverage and that the femtocell coverage does not leak into an outdoor macrocell.
△ Less
Submitted 14 October, 2010; v1 submitted 12 October, 2009;
originally announced October 2009.
