Leading or working on View-through rate modeling, Conversion modeling, good-click modeling with attention signals and ad-integrity project (UBP suspension), launched 10+ model in production, onboarding new product, ranging from feature engineering/label engineering/feature embedding, deep model architecture improvement, to training/serving optimization/acceleration, generate x$ million revenue with significant user engagement improvement and advertiser ROI improvement.

leading the project of “advertising forecasting and bid recommendation for native and search advertisement”

– building CTR/CVR forecasting, Bid landscape model, bid recommendation model, supply forecasting model, targeting forecasting models using machine learning and optimization for campaigns with different targeting (CPC, CPI, CPA campaigns, etc).

– Machine learning and optimization for smart pricing, budget recommendation, targeting recommendation, portfolio optimization, etc

– Deep learning for advertisement (Attention, CNN, heterogeneous embedding, etc)

• Code in production, publications in AAAI'18, WWW'18x

• Primary leader for: "Science-driven innovations for Mobile AI: Algorithm and Applications" . link: https://arxiv.org/pdf/1711.07580.pdf

* Highlight:

(i) Bring algorithm-level innovation (including deep learning, optimization) and concept-level innovation (more statistics) into mobile data science;

(ii) 2015 Samsung best paper award (the only one in Samsung U.S)

(iii) 1st Samsung deep learning paper published in top conference; the most prolific AI author at Samsung

(iv) Prototype and code transfer: Prototypes in PUA (potentially unwanted apps) and malice, image privacy;

(v) 10+ papers in AAAI, WSDM, SDM, CIKM and CCS, ACSAC, DSN, etc; Several patents filed

* Thrust 1: Mobile App Risk Assessment:

(i) App security risk from code perspective: permission, ads, API, call graph.

(ii) meta-data perspective: semantic understanding from word-level embedding to sentence-level and app-level embedding;

(iii) malicious app detection: contextual API call graph; novel attack models

* Thrust 2: Mobile Recommendation and Targeting

(i) app recommendation based on factorized machine taking into external factors

(ii) context aware rec: Tensor Factorization

* Thrust 3: Image Privacy and Porn image detection for Mobile

(i) Photo privacy protection using CNN and RNN

(ii) Multi-view feature extraction for sensitive object detection;

(iii) New concept: "Personalized image privacy" by enforcing photo sharing patterns.

(iv) Cloud side privacy enhancement using image perturbation

(v) Porn Image detection using deep features and architecture

* Thrust 4: Deep Learning on Mobile Device

(i) Acceleration, model adaption and enhancement

* Thrust 5: Bayesian Causal impact analysis and User Profiling/Targeting for Mobile Apps

(i) Ad-hoc and Bayesian casual inference, propensity score

(ii) Proposals and (to do) works on user targeting: targeting model, look-alike model, etc.

* Led a project about "Exclusive Sparsity for robust feature learning".

(i) designed structural sparsity learning algorithms for feature selection based on new models;

(ii) proposed iteratively reweighted (EM-like) algorithms for a class of "exclusive LASSO" problem;

(iii) applied the method to solve multi-view image annotation problems, bio-mark identification and multi-task uncorrelated feature learning problems.

See NIPS, AAAI papers.