-
Eagle: End-to-end Deep Reinforcement Learning based Autonomous Control of PTZ Cameras
Authors:
Sandeep Singh Sandha,
Bharathan Balaji,
Luis Garcia,
Mani Srivastava
Abstract:
Existing approaches for autonomous control of pan-tilt-zoom (PTZ) cameras use multiple stages where object detection and localization are performed separately from the control of the PTZ mechanisms. These approaches require manual labels and suffer from performance bottlenecks due to error propagation across the multi-stage flow of information. The large size of object detection neural networks al…
▽ More
Existing approaches for autonomous control of pan-tilt-zoom (PTZ) cameras use multiple stages where object detection and localization are performed separately from the control of the PTZ mechanisms. These approaches require manual labels and suffer from performance bottlenecks due to error propagation across the multi-stage flow of information. The large size of object detection neural networks also makes prior solutions infeasible for real-time deployment in resource-constrained devices. We present an end-to-end deep reinforcement learning (RL) solution called Eagle to train a neural network policy that directly takes images as input to control the PTZ camera. Training reinforcement learning is cumbersome in the real world due to labeling effort, runtime environment stochasticity, and fragile experimental setups. We introduce a photo-realistic simulation framework for training and evaluation of PTZ camera control policies. Eagle achieves superior camera control performance by maintaining the object of interest close to the center of captured images at high resolution and has up to 17% more tracking duration than the state-of-the-art. Eagle policies are lightweight (90x fewer parameters than Yolo5s) and can run on embedded camera platforms such as Raspberry PI (33 FPS) and Jetson Nano (38 FPS), facilitating real-time PTZ tracking for resource-constrained environments. With domain randomization, Eagle policies trained in our simulator can be transferred directly to real-world scenarios.
△ Less
Submitted 9 April, 2023;
originally announced April 2023.
-
Machine Learning for Microcontroller-Class Hardware: A Review
Authors:
Swapnil Sayan Saha,
Sandeep Singh Sandha,
Mani Srivastava
Abstract:
The advancements in machine learning opened a new opportunity to bring intelligence to the low-end Internet-of-Things nodes such as microcontrollers. Conventional machine learning deployment has high memory and compute footprint hindering their direct deployment on ultra resource-constrained microcontrollers. This paper highlights the unique requirements of enabling onboard machine learning for mi…
▽ More
The advancements in machine learning opened a new opportunity to bring intelligence to the low-end Internet-of-Things nodes such as microcontrollers. Conventional machine learning deployment has high memory and compute footprint hindering their direct deployment on ultra resource-constrained microcontrollers. This paper highlights the unique requirements of enabling onboard machine learning for microcontroller class devices. Researchers use a specialized model development workflow for resource-limited applications to ensure the compute and latency budget is within the device limits while still maintaining the desired performance. We characterize a closed-loop widely applicable workflow of machine learning model development for microcontroller class devices and show that several classes of applications adopt a specific instance of it. We present both qualitative and numerical insights into different stages of model development by showcasing several use cases. Finally, we identify the open research challenges and unsolved questions demanding careful considerations moving forward.
△ Less
Submitted 20 December, 2022; v1 submitted 28 May, 2022;
originally announced May 2022.
-
MANGO: A Python Library for Parallel Hyperparameter Tuning
Authors:
Sandeep Singh Sandha,
Mohit Aggarwal,
Igor Fedorov,
Mani Srivastava
Abstract:
Tuning hyperparameters for machine learning algorithms is a tedious task, one that is typically done manually. To enable automated hyperparameter tuning, recent works have started to use techniques based on Bayesian optimization. However, to practically enable automated tuning for large scale machine learning training pipelines, significant gaps remain in existing libraries, including lack of abst…
▽ More
Tuning hyperparameters for machine learning algorithms is a tedious task, one that is typically done manually. To enable automated hyperparameter tuning, recent works have started to use techniques based on Bayesian optimization. However, to practically enable automated tuning for large scale machine learning training pipelines, significant gaps remain in existing libraries, including lack of abstractions, fault tolerance, and flexibility to support scheduling on any distributed computing framework. To address these challenges, we present Mango, a Python library for parallel hyperparameter tuning. Mango enables the use of any distributed scheduling framework, implements intelligent parallel search strategies, and provides rich abstractions for defining complex hyperparameter search spaces that are compatible with scikit-learn. Mango is comparable in performance to Hyperopt, another widely used library. Mango is available open-source and is currently used in production at Arm Research to provide state-of-art hyperparameter tuning capabilities.
△ Less
Submitted 22 May, 2020;
originally announced May 2020.
-
StreetX: Spatio-Temporal Access Control Model for Data
Authors:
Sandeep Singh Sandha
Abstract:
Cities are a big source of spatio-temporal data that is shared across entities to drive potential use cases. Many of the Spatio-temporal datasets are confidential and are selectively shared. To allow selective sharing, several access control models exist, however user cannot express arbitrary space and time constraints on data attributes using them. In this paper we focus on spatio-temporal access…
▽ More
Cities are a big source of spatio-temporal data that is shared across entities to drive potential use cases. Many of the Spatio-temporal datasets are confidential and are selectively shared. To allow selective sharing, several access control models exist, however user cannot express arbitrary space and time constraints on data attributes using them. In this paper we focus on spatio-temporal access control model. We show that location and time attributes of data may decide its confidentiality via a motivating example and thus can affect user's access control policy. In this paper, we present StreetX which enables user to represent constraints on multiple arbitrary space regions and time windows using a simple abstract language. StreetX is scalable and is designed to handle large amount of spatio-temporal data from multiple users. Multiple space and time constraints can affect performance of the query and may also result in conflicts. StreetX automatically resolve conflicts and optimizes the query evaluation with access control to improve performance. We implemented and tested prototype of StreetX using space constraints by defining region having 1749 polygon coordinates on 10 million data records. Our testing shows that StreetX extends the current access control with spatio-temporal capabilities.
△ Less
Submitted 10 November, 2017;
originally announced November 2017.
-
Complex Event Processing of Health Data in Real-time to Predict Heart Failure Risk and Stress
Authors:
Sandeep Singh Sandha,
Mohammad Kachuee,
Sajad Darabi
Abstract:
In this paper, we develop a scalable system which can do real-time analytics for different health applications. The occurrence of different health conditions can be regarded as the complex events and thus this concept can be extended to other use cases easily. Large number of users should be able to send the data in real-time, and should be able to receive the feedback and result. Kee** the requ…
▽ More
In this paper, we develop a scalable system which can do real-time analytics for different health applications. The occurrence of different health conditions can be regarded as the complex events and thus this concept can be extended to other use cases easily. Large number of users should be able to send the data in real-time, and should be able to receive the feedback and result. Kee** the requirements in mind we used Kafka and Spark to develop our system. In this setting, multiple users are running Kafka producer clients, which are sending data in real-time. Spark streaming is used to process data from Kafka of different window sizes to analyze the health conditions. We have developed and tested the heart attack risk and stress prediction as our sample complex events detection use cases. We have simulated and tested our system with multiple health datasets.
△ Less
Submitted 13 July, 2017;
originally announced July 2017.
-
The GangaWatch Mobile App to Enable Usage of Water Data in Every Day Decisions Integrating Historical and Real-time Sensing Data
Authors:
Sandeep S Sandha,
Biplav Srivastava,
Sukanya Randhawa
Abstract:
We demonstrate a novel mobile application called GangaWatch that makes water pollution data usable and accessible focusing on one of the most polluted river basins in the world. It is intended to engage common public who want to see water condition and safe limits, and their relevance based on different purposes. The data is a combination of old data determined from lab tests on physical samples a…
▽ More
We demonstrate a novel mobile application called GangaWatch that makes water pollution data usable and accessible focusing on one of the most polluted river basins in the world. It is intended to engage common public who want to see water condition and safe limits, and their relevance based on different purposes. The data is a combination of old data determined from lab tests on physical samples and new data from real time sensors collected from Ganga basin. The platform is open for contribution from others, the data is also available for reuse via public APIs, and it has already been used to derive new insights.
△ Less
Submitted 27 January, 2017;
originally announced January 2017.
