scholarly journals Evaluating Federated Learning Scenarios in a Tumor Classification Application

2021 ◽  
Author(s):  
Rafaela C. Brum ◽  
George Teodoro ◽  
Lúcia Drummond ◽  
Luciana Arantes ◽  
Maria Clicia Castro ◽  
...  
Keyword(s):  
Machine Learning ◽  
Execution Time ◽  
Data Privacy ◽  
Tumor Infiltrating Lymphocytes ◽  
Tumor Classification ◽  
Financial Cost ◽  
Privacy Concerns ◽  
Learning Scenarios ◽  
Infiltrating Lymphocytes ◽  
Distributed Machine Learning

Federated Learning is a new area of distributed Machine Learning (ML) that emerged to deal with data privacy concerns. In this approach, each client has access to a local and private dataset. They only exchange the model weights and updates. This paper presents a Federated Learning (FL) approach to a cloud Tumor-Infiltrating Lymphocytes (TIL) application. The results show that the FL approach outperformed the centralized one in all evaluated ML metrics. It also reduced the execution time although the financial cost has increased.

scholarly journals Communication-efficient and Scalable Decentralized Federated Edge Learning

2021 ◽  
Author(s):  
Austine Zong Han Yapp ◽  
Hong Soo Nicholas Koh ◽  
Yan Ting Lai ◽  
Jiawen Kang ◽  
Xuandi Li ◽  
...  
Keyword(s):  
Machine Learning ◽  
Data Privacy ◽  
Large Scale ◽  
Local Model ◽  
Communication Overhead ◽  
Model Training ◽  
Collaborative Training ◽  
Distributed Machine Learning ◽  
The Coordinator

Federated Edge Learning (FEL) is a distributed Machine Learning (ML) framework for collaborative training on edge devices. FEL improves data privacy over traditional centralized ML model training by keeping data on the devices and only sending local model updates to a central coordinator for aggregation. However, challenges still remain in existing FEL architectures where there is high communication overhead between edge devices and the coordinator. In this paper, we present a working prototype of blockchain-empowered and communication-efficient FEL framework, which enhances the security and scalability towards large-scale implementation of FEL.

Computerized features of spatial arrangement of tumor-infiltrating lymphocytes from H&E images predicts survival and response to checkpoint inhibitors in gynecologic cancers.

2020 ◽  
Vol 38 (15_suppl) ◽  
pp. 6074-6074
Author(s):  
Sepideh Azarianpour Esfahani ◽  
Germán Corredor ◽  
Kaustav Bera ◽  
PingFu Fu ◽  
Amy Joehlin-Price ◽  
...  
Keyword(s):  
Machine Learning ◽  
Cox Regression ◽  
Checkpoint Inhibitors ◽  
Progression Free Survival ◽  
Tumor Infiltrating Lymphocytes ◽  
Spatial Arrangement ◽  
Ovarian Carcinomas ◽  
Gynecologic Cancers ◽  
Prognostic Features ◽  
Infiltrating Lymphocytes

6074 Background: Immune checkpoint inhibitors (ICI) have demonstrated success in solid tumors. In gynecologic cancers (GC), the response rate is still low (~10-15%) except in MSI-H endometrial cancer (~ 50%). Current biomarkers (e.g. PDL1 expression) have limited utility in identifying benefit from ICI in GC. In this work we evaluated the ability of computational measurements of spatial arrangement of tumor infiltrating lymphocytes (TIL) from H&E slide images in predicting overall survival (OS) and response to ICI in ovarian, cervical and endometrial cancers. Methods: The study included 151 patients, including 102 ovarian carcinomas treated with surgery and chemotherapy (D1) and another set (D2) of n=49 patients (n=14 ovarian, n=27 endometrial and n=8 cervical), treated with different ICI agents (Pembrolizumab, Nivolumab, Ipilimumab, Avelumab) in the second line setting. Progressors and non-progressors in D2 were classified according to clinical improvement and radiologic assessment by RECIST. A machine learning approach was employed to identify tumor regions on the diagnostic slides from D1 and D2 and then used to automatically identify TILs within the tumor regions. Subsequently machine learning was used to define TIL clusters based on TIL proximity, and graph network theory was used to capture measurements relating to spatial arrangement of TIL clusters. The multivariable Cox regression model (MCRM) was trained on n=51 patients from D1 to predict OS and then independently evaluated in predicting (1) OS on the hold-out n=51 patients in D1 and (2) response and progression-free survival (PFS) in D2. Results: Statistical analysis identified 7 prognostic features relating to interaction of TIL clusters with cancer nuclei. MCRM was prognostic of OS on the n=51 hold out patients in D1 (hazard ratio (HR)=2.06, 95% confidence interval [1.04- 4.07], p=0.008) and predictive of PFS in D2 (HR=2.24, CI=[1.13-4.44], p=0.03). The AUC for MCRM in predicting progression in D2 was 82%. Conclusions: Computerized features of spatial arrangement of TILs on H&E images were prognostic of OS and PFS and predicted response to ICI in three gynecological cancers. These findings need to be validated in larger, multi-site validation sets. [Table: see text]

scholarly journals XFinder: Detecting Unknown Anomalies in Distributed Machine Learning Scenario

2021 ◽  
Vol 3 ◽  
Author(s):  
Haizhou Du ◽  
Shiwei Wang ◽  
Huan Huo
Keyword(s):  
Machine Learning ◽  
Anomaly Detection ◽  
Real Time ◽  
Network Traffic ◽  
Security And Privacy ◽  
Closed Sets ◽  
Online Mode ◽  
Average Accuracy ◽  
Learning Scenarios ◽  
Distributed Machine Learning

In recent years, the emergence of distributed machine learning has enabled deep learning models to ensure data security and privacy while training efficiently. Anomaly detection for network traffic in distributed machine learning scenarios is of great significance for network security. Although deep neural networks have made remarkable achievements in anomaly detection for network traffic, they mainly focus on closed sets, that is, assuming that all anomalies are known. However, in a real network environment, unknown abnormalities are fatal risks faced by the system because they have no labels and occur before the known anomalies. In this study, we design and implement XFinder, a dynamic unknown traffic anomaly detection framework in distributed machine learning. XFinder adopts an online mode to detect unknown anomalies in real-time. XFinder detects unknown anomalies by the unknowns detector, transfers the unknown anomalies to the prior knowledge base by the network updater, and adopts the online mode to report new anomalies in real-time. The experimental results show that the average accuracy of the unknown anomaly detection of our model is increased by 27% and the average F1-Score is improved by 20%. Compared with the offline mode, XFinder’s detection time is reduced by an average of approximately 33% on three datasets, and can better meet the network requirement.

scholarly journals Distributed Machine Learning through Heterogeneous Edge Systems

2020 ◽  
Vol 34 (05) ◽  
pp. 7179-7186
Author(s):  
Hanpeng Hu ◽  
Dan Wang ◽  
Chuan Wu
Keyword(s):  
Machine Learning ◽  
Convergence Time ◽  
Privacy Concerns ◽  
Model Update ◽  
Large Heterogeneity ◽  
Testbed Implementation ◽  
Model Training ◽  
Iot Devices ◽  
Core Idea ◽  
Distributed Machine Learning

Many emerging AI applications request distributed machine learning (ML) among edge systems (e.g., IoT devices and PCs at the edge of the Internet), where data cannot be uploaded to a central venue for model training, due to their large volumes and/or security/privacy concerns. Edge devices are intrinsically heterogeneous in computing capacity, posing significant challenges to parameter synchronization for parallel training with the parameter server (PS) architecture. This paper proposes ADSP, a parameter synchronization model for distributed machine learning (ML) with heterogeneous edge systems. Eliminating the significant waiting time occurring with existing parameter synchronization models, the core idea of ADSP is to let faster edge devices continue training, while committing their model updates at strategically decided intervals. We design algorithms that decide time points for each worker to commit its model update, and ensure not only global model convergence but also faster convergence. Our testbed implementation and experiments show that ADSP outperforms existing parameter synchronization models significantly in terms of ML model convergence time, scalability and adaptability to large heterogeneity.

scholarly journals Clinical Significance and Molecular Annotation of Cellular Morphometric Subtypes in Lower Grade Gliomas discovered by Machine Learning: a retrospective multicentric study

2021 ◽  
Author(s):  
Xiao-Ping Liu ◽  
Xiaoqing Jin ◽  
Saman Ahmadian ◽  
Xu Yang ◽  
Su-Fang Tian ◽  
...  
Keyword(s):  
Machine Learning ◽  
Poor Prognosis ◽  
Tumor Infiltrating Lymphocytes ◽  
Lower Grade ◽  
Primary Therapy ◽  
Multicentric Study ◽  
Molecular Alterations ◽  
The Poor ◽  
Infiltrating Lymphocytes ◽  
Checkpoint Genes

Abstract Lower grade gliomas (LGGs) are heterogenous diseases by clinical, histological and molecular criteria. Here, we developed a machine learning pipeline to extract cellular morphometric biomarkers from whole slide images of tissue histology; and identified and externally validated robust cellular morphometric subtypes of LGGs in multi-center cohorts. The subtypes have significantly independent predictive power for overall survival across all three independent cohorts. In the TCGA-LGG cohort, we found that patients within the poor-prognosis subtype responded poorly to primary therapy and follow-up treatment. Furthermore, LGGs within the poor-prognosis subtype were characterized by higher mutational burden, higher frequencies of copy number alterations, and higher level of tumor-infiltrating lymphocytes and immune checkpoint genes. Higher level of PD-1/PD-L1/CTLA-4 was confirmed by immunohistochemical staining. In addition, the subtypes learned from LGG demonstrates translational impact on Glioblastoma (GBM). Overall, we developed and validated a framework for the cellular morphometric subtype discovery in LGGs associated with specific molecular alterations, immune micro-environment, prognosis and treatment response.

scholarly journals Clinical Significance and Molecular Annotation of Cellular Morphometric Subtypes in Lower Grade Gliomas discovered by Machine Learning: a retrospective multicentric study

2021 ◽  
Author(s):  
Xiao-Ping Liu ◽  
Xiaoqing Jin ◽  
Saman Ahmadian ◽  
Xu Yang ◽  
Su-Fang Tian ◽  
...  
Keyword(s):  
Machine Learning ◽  
Poor Prognosis ◽  
Tumor Infiltrating Lymphocytes ◽  
Lower Grade ◽  
Primary Therapy ◽  
Multicentric Study ◽  
Molecular Alterations ◽  
The Poor ◽  
Infiltrating Lymphocytes ◽  
Checkpoint Genes

Abstract Lower grade gliomas (LGGs) are heterogenous diseases by clinical, histological and molecular criteria. Here, we developed a machine learning pipeline to extract cellular morphometric biomarkers from whole slide images of tissue histology; and identified and externally validated robust cellular morphometric subtypes of LGGs in multi-center cohorts. The subtypes have significantly independent predictive power for overall survival across all three independent cohorts. In the TCGA-LGG cohort, we found that patients within the poor-prognosis subtype responded poorly to primary therapy and follow-up treatment. Furthermore, LGGs within the poor-prognosis subtype were characterized by higher mutational burden, higher frequencies of copy number alterations, and higher level of tumor-infiltrating lymphocytes and immune checkpoint genes. Higher level of PD-1/PD-L1/CTLA-4 was confirmed by immunohistochemical staining. In addition, the subtypes learned from LGG demonstrates translational impact on Glioblastoma (GBM). Overall, we developed and validated a framework for the cellular morphometric subtype discovery in LGGs associated with specific molecular alterations, immune micro-environment, prognosis and treatment response.

scholarly journals A Comprehensive Survey of Privacy-preserving Federated Learning

2021 ◽  
Vol 54 (6) ◽  
pp. 1-36
Author(s):  
Xuefei Yin ◽  
Yanming Zhu ◽  
Jiankun Hu
Keyword(s):  
Machine Learning ◽  
Data Privacy ◽  
Rapid Development ◽  
Privacy Preserving ◽  
Future Research ◽  
Research Directions ◽  
Privacy Concerns ◽  
Privacy Leakage ◽  
Comprehensive Survey ◽  
Future Research Directions

The past four years have witnessed the rapid development of federated learning (FL). However, new privacy concerns have also emerged during the aggregation of the distributed intermediate results. The emerging privacy-preserving FL (PPFL) has been heralded as a solution to generic privacy-preserving machine learning. However, the challenge of protecting data privacy while maintaining the data utility through machine learning still remains. In this article, we present a comprehensive and systematic survey on the PPFL based on our proposed 5W-scenario-based taxonomy. We analyze the privacy leakage risks in the FL from five aspects, summarize existing methods, and identify future research directions.

Sign in / Sign up

Export Citation Format

Share Document