Provable Privacy Guarantee, and, Smart Navigation to Aid Glaucoma Patients
By Syed Hafiz, Phd, "Enabling Advanced Queries on Untrusted Databases with Provable Privacy Guarantee and Superior Efficiency",
By Reeva Patel, Gauri Todur, Nidhi Thankasala, "Novel Machine Learning-Based Smart Navigation Attachment to Aid Glaucoma Patients"
Abstract: Privacy is a fundamental human right. Service providers can track our queries and interests when we search for (or consume) any information on social media, video-streaming sites, public databases, etc. They can infer personally identifiable and sensitive information like political preference, sexual orientation, etc., which users may not like to share. Private information retrieval (PIR) is a powerful cryptographic primitive that solves this ubiquitous problem of safeguarding the privacy of users' access patterns to remote, untrusted databases. Most PIR techniques only support position-based queries – which require the knowledge of the data record or block index – to access the databases. However, in the real world, users are more interested in retrieving records by various search criteria, e.g., searching by keywords, sorting and TopK search, aggregate queries, and ultimately, SQL-like utility. To protect users' sensitive access patterns on untrusted databases and support regular and advanced queries, e.g., aggregate and SQL queries, we need new PIR techniques that provide provable security guarantee and efficiency in terms of computation and communication complexity associated with PIR tasks.
Short-bio: Syed Hafiz is an Intel Noyce Postdoctoral Fellow at the University of California, Davis. His current research focuses on security and privacy issues of deep neural networks (DNNs) and cryptography, e.g., secure multi-party computation (MPC)-based privacy-preserving machine learning (PPML). Very recently, his team exposed two side-channel vulnerabilities (e.g., privacy and confidentiality) of DNN models to reverse-engineer the label of the unknown input image (DATE 2022) utilizing cache-based microarchitectural traces and fingerprint the family of the victim model (EuroS&P 2022), collecting global system traces. He completed his Ph.D. in Computer Science from Indiana University-Bloomington. He worked on transitioning Private Information Retrieval (PIR), a powerful privacy-preserving cryptographic primitive, from a theoretical construct to a valuable tool in the privacy practitioners' toolkit. He introduced a novel method called "indexes of queries" that allow users to query PIR databases using so-called "expressive'' queries (ACM CCS 2017). Also, he presented new protocols that exhibit "essentially optimal'' efficiency – both in practice and in theory – for the class of so-called "1-private'' PIR constructions (PETS 2019). Besides, he instructed an undergrad crypto course, CSCI-C231: Introduction to the Mathematics of Cybersecurity, at IU CS for three consecutive semesters.
"Smart Navigation Attachment to Aid" Abstract:
Glaucoma, a serious eye condition, is the second leading cause of blindness in the world, and currently affects 80 million people. Its irreversibility necessitates the use of navigation aids, which are becoming more tech-oriented. However, current smart mobility aids are very costly and are only utilized by 2-6% of the visually impaired population.
We analyzed various existing mobility aids and created a list of successful and necessary components. Our engineering goal is to create a product that mounts onto any size-diameter white cane, detects and recognizes any object, quickly and alerts the user through speakers. With a list of many features, while maintaining a relatively low cost of under $250. Our average accuracy overall was 80.4%. After iterating our design, we calculated our final overall accuracy as 90.0%. We successfully met our criteria of minimum 85% overall accuracy. Our product can help a wider range of visually impaired people navigate the outdoors more safely, efficiently, and cost-effectively.
Video Demo: https://youtu.be/aQUawlxM2vo
Bio:
We are a team from the Stem Leadership Institute at Cabrillo Middle School, and we are passionate about assisting our community. We developed this device because we were motivated by a personal connection and wanted to make a difference. Gauri was enthusiastic about programming and took on the many coding elements needed. Reeva was driven by building the device and helped design and construct the first prototypes. Finally, Nidhi was devoted to researching and testing; she recorded data and made high-level graphs.
0:00 Chapter Introduction
4:46 Navigation Aid Intro
6:26 Navigation Aid Presentation
24:10 Navigation Aid Q&A
25:35 Privacy Guarantee Intro
26:17 Privacy Guarantee Presentation
1:14:10 Privacy Guarantee Q&A
Видео Provable Privacy Guarantee, and, Smart Navigation to Aid Glaucoma Patients канала San Francisco Bay ACM
By Reeva Patel, Gauri Todur, Nidhi Thankasala, "Novel Machine Learning-Based Smart Navigation Attachment to Aid Glaucoma Patients"
Abstract: Privacy is a fundamental human right. Service providers can track our queries and interests when we search for (or consume) any information on social media, video-streaming sites, public databases, etc. They can infer personally identifiable and sensitive information like political preference, sexual orientation, etc., which users may not like to share. Private information retrieval (PIR) is a powerful cryptographic primitive that solves this ubiquitous problem of safeguarding the privacy of users' access patterns to remote, untrusted databases. Most PIR techniques only support position-based queries – which require the knowledge of the data record or block index – to access the databases. However, in the real world, users are more interested in retrieving records by various search criteria, e.g., searching by keywords, sorting and TopK search, aggregate queries, and ultimately, SQL-like utility. To protect users' sensitive access patterns on untrusted databases and support regular and advanced queries, e.g., aggregate and SQL queries, we need new PIR techniques that provide provable security guarantee and efficiency in terms of computation and communication complexity associated with PIR tasks.
Short-bio: Syed Hafiz is an Intel Noyce Postdoctoral Fellow at the University of California, Davis. His current research focuses on security and privacy issues of deep neural networks (DNNs) and cryptography, e.g., secure multi-party computation (MPC)-based privacy-preserving machine learning (PPML). Very recently, his team exposed two side-channel vulnerabilities (e.g., privacy and confidentiality) of DNN models to reverse-engineer the label of the unknown input image (DATE 2022) utilizing cache-based microarchitectural traces and fingerprint the family of the victim model (EuroS&P 2022), collecting global system traces. He completed his Ph.D. in Computer Science from Indiana University-Bloomington. He worked on transitioning Private Information Retrieval (PIR), a powerful privacy-preserving cryptographic primitive, from a theoretical construct to a valuable tool in the privacy practitioners' toolkit. He introduced a novel method called "indexes of queries" that allow users to query PIR databases using so-called "expressive'' queries (ACM CCS 2017). Also, he presented new protocols that exhibit "essentially optimal'' efficiency – both in practice and in theory – for the class of so-called "1-private'' PIR constructions (PETS 2019). Besides, he instructed an undergrad crypto course, CSCI-C231: Introduction to the Mathematics of Cybersecurity, at IU CS for three consecutive semesters.
"Smart Navigation Attachment to Aid" Abstract:
Glaucoma, a serious eye condition, is the second leading cause of blindness in the world, and currently affects 80 million people. Its irreversibility necessitates the use of navigation aids, which are becoming more tech-oriented. However, current smart mobility aids are very costly and are only utilized by 2-6% of the visually impaired population.
We analyzed various existing mobility aids and created a list of successful and necessary components. Our engineering goal is to create a product that mounts onto any size-diameter white cane, detects and recognizes any object, quickly and alerts the user through speakers. With a list of many features, while maintaining a relatively low cost of under $250. Our average accuracy overall was 80.4%. After iterating our design, we calculated our final overall accuracy as 90.0%. We successfully met our criteria of minimum 85% overall accuracy. Our product can help a wider range of visually impaired people navigate the outdoors more safely, efficiently, and cost-effectively.
Video Demo: https://youtu.be/aQUawlxM2vo
Bio:
We are a team from the Stem Leadership Institute at Cabrillo Middle School, and we are passionate about assisting our community. We developed this device because we were motivated by a personal connection and wanted to make a difference. Gauri was enthusiastic about programming and took on the many coding elements needed. Reeva was driven by building the device and helped design and construct the first prototypes. Finally, Nidhi was devoted to researching and testing; she recorded data and made high-level graphs.
0:00 Chapter Introduction
4:46 Navigation Aid Intro
6:26 Navigation Aid Presentation
24:10 Navigation Aid Q&A
25:35 Privacy Guarantee Intro
26:17 Privacy Guarantee Presentation
1:14:10 Privacy Guarantee Q&A
Видео Provable Privacy Guarantee, and, Smart Navigation to Aid Glaucoma Patients канала San Francisco Bay ACM
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
Faster Data Science+Best Practices for AI-enabled applications, Ian Swanson&Rama Akkiraju 20180827A Smarter Process for Sensing the Information Space - Scott Spangler , IBM 20101122CHOMP Mobile App Search, by Cathy Edwards, 2011-11-28COVID-19 Conundrum: Critical Insights on Cloud Security Posture of Enterprises, Matt ChiodiNanobody Optimization via Machine LearningEmerging AI Applications & Infrastructure Systems, Dr. Wen-mei W. Hwu,20190522Enterprise AI: The most untapped opportunity for novel and impactful AI work2024 Engineers' Week Banquet and Hall of Fame CeremonyTaming Latency Variability and Scaling Deep Learning, by Jeff Dean, Google, 20131016ACM Data Mining Camp at eBay on October 15 2011 (Promo video)A new neural network for optimal time series processingHeuristic Design of Experiments with Meta-Gradient Search of Model by Greg Makowski 20110228A Big Data Solution using JSONiq, by Dr. Matthias Brantner 20130826Panel: Practical Realities of ConsultingScalable Recommender/Advertising Systems, Speaker Joaquin A. Delgado, PhD. 20130923Technology Applications for Seismic Events, Jennifer Strauss 20140917Deep Learning on Apache Spark w/ BigDL, Sergey V Ermolin , 20170626A Primer on the Major Topics in AI Law and PolicyWhen Science Meets Data: Three Stories from the Field,Speaker: Neel Sundaresan, eBAY, 20130325Knowledge Graph Extraction from Unstructured Data and Semantic Role Labeling