Eric Wu Email & Phone Number

Los Altos, Ca, Us

Palo Alto, Ca, Us

Lead the AI team in Docomo Innovations. Managing AI team in technical perspective. Direct/lead development/research of AI related projects. Proficient level PyTorch, basic Tensorflow. Subjects included (but not limited to) the following:(1) Deep Learning based 2D image recognition/classification/segmentation/generation/super-resolution. Deep 2D Vision based manufacturing assembly line defect detection/measurements. Deep learning based 3D representation/compression/detection/Mesh completion. 3D (implicit/nerf ) construction with 2D supervision.(2) Most recent CNN/RNN/Transformer (linear complexity variants)/GAN research & implementations with applications in content creation. (3) Representation learning/self-supervised pre-training from images/wireless data/user data. Special data/label architecture (GNN, various semi-supervised scenarios, labels with hierarchy, labels with ordinal requirements, data with substructures as sets). (4) Chinese/English NLP, general data analytics with Machine Learning and Deep Learning. Deep Learning based text extraction from HTML, semantic search result re-ranking, relation extraction from free text. User behavior data analysis and predictions, text summarization. Text clusters label generation. Recent language models (BERT, GPT) and their applications. LLM enhanced features for various recommendation problems.(5) Open source LLM (Llama2 based), PEFT, LDM (UNet & Transformer based), SD for 3D Object construction/3D scene/synthetic HPE data generation. LoRA tuning for SD/LLM finetuning with catastrophic forgetting reduction. AI cost reduction (pruning/various quantization/MOE/check point merging, SLM). Multi-modal VLM. Quantization/distillation of LLMs, (6) Data annotation pipeline design/planning/bias removal. ML & Deep Learning hyper-parameter optimization.(8) Mentoring Interns/Junior/Intermediate level researchers/engineers

Leading research team. Conduct research in AI/ML/NN/NLP, lead development of new concept prototypes. Apache Spark cluster setup and programming.

Lead a team of data scientists. Big data analytic. Machine learning and data mining algorithm design/assessment & research. Chinese AI dialog system (domain specific Siri-like) algorithm design and implementation.

Taipei City, Tw

Lead a 50+ R&D team in a Internet of Things (IOT) company spreading China, Taiwan. Built and operate Cloud and Big Data team. Responsible for all ThroughTek's product lines (p2p nat traversal, cloud).

• Build up and lead algorithm team, developed all core machine learning/datamining algorithms used in the company product lines.• Project cases: semantic search engine, Chinese keywords extraction technology, structured data extraction from web data, social network user spectrum, social media monitoring, cookie based audience targeting for advertisement, fine grained sentiment analysis, topics and themes analysis, topical keywords and trend anal- ysis, WYSIWYG Web crawler, Chinese web news clustering and summarization.• Served customers including large e-commerce, online/mobile advertisement, busi- ness intelligence, government bureau, etc.• Algorithms that can be applied to large scale: Streaming SVM, Fast algorithm for SVD (LSI), LDA, Fast NMF for text clustering and topics finding. Linear time spectral clustering for large scale text (power iteration). Locality sensitive hashing. Various random projection based sampling techniques.• Software porting: liblinear muti-processed and multithreaded version. C + + bridge with Python. StreamSVM porting to Windows.• Software package: Python 2, scikit-learn, numpy, scipy, gensim. C++ boost, eigen. Liblinear, LibFM. Lucene (with some inner working algorithms)• Research and work on web engine and crawler side: Cover both Webkit and Gecko XULrunner engine. 1. Web template clustering for Nutch style crawl. 2. WYSIWYG style webpage crawler.• Basic algorithms development: Chinese natural language processing. Chinese feature selections. Search engine result clustering. Chinese text categorization and sentiment analysis. Statistical based Chinese segmenter and name entity extraction. Textrank, sentences and keywords importance ranking and Chinese text summarization. Semi-supervised clustering.

•Programmed the major portion of a lighting control system application, with C#.NET CF (about 500000 lines of code) on WinCE6.• Lighting control algorithm/logic programming on ARM7 (LPC23xx) with Keil.• WinCE 6.0 image build up for company boards and driver tweaking.• Migrated and managed source control system (Surround SCM) for code running on new Douglas product lines.

NxNova focuses on professional on both image processing, recognition and Windows CE/Windows Mobile based software solutions dedicated to offering comprehensive and innovative software services and products, customized to meet our clients business needs.• Image Processing & Recognition algorithms design and implementation. Machine learning (neural network and SVM) research. Multi-core programming with In- tel threading building blocks. Parallel algorithms and application in imaging algorithms implementation.• Major programmer and algorithm designer for NxNova Vehicle License Plate Recognition system. Research on Iris recognition.

Managed a small team responsible for WinCE kernel and Embedded Linux kernel & drivers for all Empower product lines. Ported Android to run on Empower boards

Fortuna Electronic Corp.'s US branch office.• Fortuna Sunnyvale branch & RD team build up & training and project manage-ment.• Build up the entire software product line running on Fortuna PND.• GIS/GPS related technology algorithm (spatial database data extraction, sorting and searching, asymmetric data compression, fast practical routing and map rendering) research & implementation.

Phd, Mathematics

B.S., Mathematics