Multi class activity classification in videos using Motion History Image generation
Oct 10, 2024
Human action recognition has been a topic of interest across multiple fields ranging from security to entertainment systems. Tracking the motion and identifying the action being performed on a real time basis is necessary for critical security systems. In entertainment, especially gaming, the need for immediate responses for actions and gestures are paramount for the success of that system. We show that Motion History image has been a well established framework to capture the temporal and activity information in multi dimensional detail enabling various usecases including classification. We utilize MHI to produce sample data to train a classifier and demonstrate its effectiveness for action classification across six different activities in a single multi-action video. We analyze the classifier performance and identify usecases where MHI struggles to generate the appropriate activity image and discuss mechanisms and future work to overcome those limitations.
Classic algorithms are fair learners: Classification Analysis of natural weather and wildfire occurrences
Sep 04, 2023
Classic machine learning algorithms have been reviewed and studied mathematically on its performance and properties in detail. This paper intends to review the empirical functioning of widely used classical supervised learning algorithms such as Decision Trees, Boosting, Support Vector Machines, k-nearest Neighbors and a shallow Artificial Neural Network. The paper evaluates these algorithms on a sparse tabular data for classification task and observes the effect on specific hyperparameters on these algorithms when the data is synthetically modified for higher noise. These perturbations were introduced to observe these algorithms on their efficiency in generalizing for sparse data and their utility of different parameters to improve classification accuracy. The paper intends to show that these classic algorithms are fair learners even for such limited data due to their inherent properties even for noisy and sparse datasets.
How eBay’s New Search Feature Was Inspired By Window Shopping
Mar 31, 2023
A new feature generates customer delight by using modern computer vision techniques to drive new search paradigms through visual discovery using Machine Learning and semantic similarity based search techniques.
Engineering Behind Coded Coupons
Sep 23, 2021
An engineering and product focused article on how dual write architecture is utilized for providing scaling for the new coded coupons feature in eBay.
Terapeak Research 2.0
Mar 04, 2020
An engineering architecture focued article detailing the data and the application pipelines for building Terapeak in eBay. Terapeak is an analytics platform providing data capabilities for sellers to analyze competition, supply and demand.
Identity Management as a Relying Party for OAuth 2.0
Oct 10, 2018
A generic identity architecture for integrating multiple identity providers using OAuth 2.0.
Google Sign-In Integration for the eBay Android App
Oct 24, 2017
An engineering focused article on how Google identity has been integrated as a managed provider with eBay Identity Infrastructure.
Finite-State Machine for Single-Use Code
Aug 30, 2016
An engineering and architecture focused article on the usage of finite state machine computation model for securing the one time authentication mechanism in eBay.
Application Resiliency Using Netflix Hystrix
Sep 08, 2015
An engineering implementation of the Netflix Hystrix circuit breaker and its utilization to create self-healing systems.
Temporal Difference for Credit Assignment and Exploration of TD (λ) Family of Learning Procedures–a Reproducibility Report
Reinforcement learning is a major branch of machine learning which uses sequence of past experiences, to help a system learn and predict optimal behaviour. There are multitude of proposed learning procedures that have attempted to perform such learning. While typical prediction problems utilized the final outcome to minimize the error, Sutton in [1] explores “temporally successive predictions” to assign credit for the actions performed. Similar to a weather forecast for the weekend getting better as time progresses, TD (Temporal Difference) learning utilizes the same method of updating estimates every time step to progress towards the optimal value. This technical report discusses the background and intuition behind the original paper [1] by recreating the experiments and investigates the reproduced empirical results to compare and contrast the assumptions and findings described in the original literature.