Steven Zhang

San Jose, CA

Linkedin Q ad0amc@r.postjobfree.com Github Æ 469-***-****

SUMMARY OF QUALIFICATIONS

6+ years of experience in machine learning and big data working in a distributed computing environment. Extensive experience of development and the deployment of multiple machine learning models. Passionate to work on influential AI products that are directly used by everyday consumers. TECHNICAL SKILLS

Programming: Python, C/C++, Java, MySQL, TensorFlow, Pytorch, Numpy, Pandas, Scikit-learn, Cuda Software: Git, Docker, OpenCV, Apache Beam, MLFlow, Amazon EMR, AWS Glue, Spark, MLlib EXPERIENCE

InnoPeak Technology Palo Alto, CA, US

Machine Learning Engineer III July 2022 - present

+ Image Segmentation(Python): Lead the technical design, development and implementation of machine learning applications. High resolution segmentation data synthesis using an innovative object placement algorithm. Developed the high-resolution semantic segmentation(person and sky) algorithm that were deployed on smartphone devices. Optimized algorithm with Halide programming(11 ms on DSP). Achieved segmentation performance comparable to or surpassing that of third-party apps. Successfully delivered SDKs with C++. Optimized the algorithms with Halide programming, achieved the runtime of 11 ms on Qualcomm DSP, and successfully delivered C++ SDKs.

+ 3D Face Generation by aligning vision and language features from the Large Language Model(GPT3.5 and 3D latent stable diffusion)(Python): Generating High-Quality 3D face models from text descriptions and aligned vision features using LLMs and CLIP. Semantic Pyramid features, which encompass multi-scale information extracted from LLM, effectively control both abstract and fine-grained face features, resulting in remarkable improvements in the quality of face generations.

Protopia AI Austin, TX, US

Senior Deep Learning Engineer July 2021 - July 2022

+ Build and operate complex, end-to-end, machine learning pipelines and design production APIs and data delivery processes with AWS S3, EC2, SQS. Extensive experience of delivering and deployment of multiple machine learning models for varous task. These tasks encompass, but not limited to:

+ Text Detection and Recognition(Python):Delivered a text data obfuscation solution that generated $100k in incremental business value annually. Implemented a text detection and transformer-based recognition network in a distributed training pipeline with multi GPUs and multi nodes, achieving state-of-the-art performance in text detection and recognition with obfuscated images.

+ Email Spam Detection with obfuscated email data(Python,C++): Developed an advanced spam detection system using a transformer (BERT) based language model on transformed email data. Preprocessed the Enron- Spam dataset, implemented a customized BERT architecture, and integrated context features for improved detection. Outperformed baseline method(random forest) by 6% and ensured the transformed email data remained unintelligible to humans.

+ Privacy Preserving Semantic Segmentation(Python, AWS): Developed a privacy-preserving image obfusca- tion module for semantic segmentation in medical imaging. The module seamlessly integrates with pretrained models, safeguarding sensitive data without compromising segmentation performance. The segmentation UNet with obfuscation module achieved 0.84 mIoU on PancreasCT dataset. Dolphin Geophysical Houston, TX, US

Signal Processing Geophysicist(Python) Aug 2015 - Nov 2017

+ Championed a $1M project for BP with marine data processing and image processing, de-noise, de-multiples, velocity model building with linear regression and non-linear inversion.

+ Involved in the development of geophysical data processing software, which is widely used by a number of oil companies such as ExxonMobil.

Sigma Cubed Inc Houston, TX, US

Junior Data Scientist(Python) Feb 2014 - May 2015

+ Increased operation revenue by $200k+/year by accomplishing 5+ micro-seismic data processing projects, utilizing micro-seismic data processing, locate micro-seismic events and fracture.

+ Velocity model building with linear regression and non-linear inversion. ACADEMIC PROJECTS

Aspect Extraction and Aspect based Sentiment Classification(Python, Pytorch)

+ Aspect extraction on restaurant and laptop dataset with finetuning pretrained Distilbert model.

+ Aspect based Sentiment classification by fine-tuning Bert model achieved 78% accuracy on restaurant dataset and 76% on laptop dataset.

Context based Question Answering

+ Fine-tune Bert on SQUAD dataset

+ Evaluate fine-tuned Bert model and Bert Squad2 with Rouge1, Rough2 and Rough-l metrics. Context based Question Answering

+ Fine-tune Bert on SQUAD dataset

+ Evaluate fine-tuned Bert with Rouge1, Rough2 and Rough-l metrics.

+ build a system taking SQUAD dataset as input to perform supervised learning, load document data and convert to Jason format, split the data into data chunks, save the embeddings in vectorstores for the later usage to interact with LLM and solve the question answering retrieval problems.

+ the predictions results from LLM system achieves much better Rough2 and Rough3 results (%3-5%) compared with finetuned Bert model.

Name Entity Recognition

+ Genearte entity with Standord NER Tagger on NER data from Kaggle.

+ Train Bidirectional LSTM for entity recognition on Annotated Corpus, achieved 0.99 accuracy. EDUCATION

University of Central Florida Orlando, FL, US

Research Assistant @CRCV Aug. 2019 - 2020

University of Texas at Arlington Arlington, TX, US Master of Computer Science GPA:4.0/4.0 Aug. 2017 - May 2019 Jilin University Changchun, Jilin, China

Bachelor of Engineering

Patent

A Data Efficient Lens Dirtiness Detection Method for Smartphone Cameras, X. Zhang et al., 2023. Publications

X.Zhang, R. Gupta, A. Mian, N. Rahnavard and M. Shah, "Cassandra: Detecting Trojaned Networks from Adversarial Perturbations," in IEEE Access, doi:10.1109/ACCESS.2021.3101289. X. Zhang, Z. Xu, S. Wang F. Zhu et.al., "Seq3seq Fingerprint: Towards End-to-end Semi-supervised Deep Drug Discovery", The 9th ACM-BCB conference, Washington DC,2018.

Contact this candidate