Yi Hong Email and Phone Number

+ designed anomaly detection and root cause analysis algorithms of AppD Cognition Engine, which achieved both false alarm rate and miss alarm rate below 20%, and the accuracy for root cause analysis above 60%Cognition Engine is AppD's first machine learning feature into production. Here are some videos about AppD Cognition Engine: https://www.youtube.com/watch?v=i63oqSyCPvohttps://www.youtube.com/watch?v=_QSD_hBh3dc+ designed and implemented AppD automated transaction diagnostics capabilities, which are capable of drilling down root cause to code level (private-beta before GA release)+ introduced fast weekly baseline and designed an algorithm of learning fast weekly baselines for baselining rapidly changing applications / systems like K8s+ designed and implemented algorithms for machine log summarizing, event (log, error message, stack trace) based anomaly detection, decision tree root cause analysis. Descriptions about my work on log summarizing and root cause analysis can be found from linkshttps://priorart.ip.com/IPCOM/000253814https://priorart.ip.com/IPCOM/000253806

+ enhanced Sumo Logic Log-Reduce feature by introducing the weighted min-hash technique for rapidly clustering machine logs into groups. Here is the description about the technique: https://www.sumologic.com/blog/rapid-similarity-search-with-weighted-min-hash/https://dzone.com/articles/rapid-similarity-search-with-weighted-min-hash+ studied and implemented many algorithms for time series modeling, anomaly detection, and prediction+ built sumo logic experimental framework with two other engineers, which was used for experimenting new machine learning algorithms on real customer machine data like logs and metrics

+ built Walmart E-Commerce first version Recommend Engine that generates recommendations for over 10 millions of customers per day including several real-time event-triggered campaigns+ built item semantic similarity graph from item catalog information such as name, brand, description, category, and image+ introduced a robust graph Laplacian smoothing algorithm (RGLS) for CTR prediction+ implemented and played with many recommendation algorithms such as item kNN, user kNN, matrix factorization and deep learning

machine learning, computer vision, statistical modeling

data clustering, feature selection, estimation of distribution algorithms