Wei Wu

Operators Library - Implemented a few basic operators for CPU, CUDA, and Cambricon MLU. - Generalized elementwise operators infrastructures for Cambricon MLU. - Optimized Cambricon elementwise infrastructures to ~50% faster than PyTorch (add, swiglu, ...). - Generalized reduce operators infrastructures for Cambricon MLU. - Optimized Cambricon reduce infrastructures to ~80% faster than PyTorch (rms norm, causal softmax, ...). - Implemented a general purpose rearrange implementation for Cambricon MLU. - Optimized a specific Cambricon rearrange implementation which improved the overall average decode speed of our inference framework by 20 ms/token. - Implemented random sample operator for Cambricon MLU. - Implemented rotary positional embedding operator for Cambricon MLU, 95% faster than PyTorch. - Fixed an obscure bug in a previous Cambricon rms norm implementation that caused precision issues in text generation. Inference Frameworks - Memory pool - Single/Multi-device inference on CPU, Nvidia GPU, Ascend NPU, and Cambricon MLU. - Web API - LRU Cache for operator descriptors. - Modualization using shared operator descriptors and workspace across rounds of inference, while also exposing simple and clean operator calls for easier inference construction. - Inference procedure implementation for gptq-int8 models. Training Frameworks AI Summer Camp Tutor

Product Exploration and Development: Explored and evaluated roadmaps for hardware and software products, driving innovation and product alignment with market needs. Designed and developed demonstrative prototypes to showcase novel ideas and concepts. Developer and Community Support: Provided technical support to the global developer community, addressing developer requests and aiding in the delivery of numerous applications and demos (e.g., DB Creations, Forward Gaming). Actively supported community operations through platforms like Discord and Reddit. Game Development for XREAL Glasses: Developed engaging games such as Mesh Jump, Karting, and Immovable, showcasing the capabilities of augmented reality hardware. Event Operations and Prototyping: Provided full-stack onsite operations for global events, including CES 2024, MWC Shanghai 2024, WAIC Shanghai 2024, and China Joy 2024. Mentored and judged at MIT Reality Hack 2024, fostering innovation in XR development. Provided prototyping and onsite support for major collaborators like Google AndroidXR. Business and Technical Support: Delivered technical support during business discussions with international stakeholders in the US, enabling partnerships and product announcements (e.g., Xbox, NVIDIA, AWE 2024). Walked through and analyzed the process of T-Mobile 5G certifications, contributing to product readiness for cutting-edge networks. Feature Development and Tooling: Developed scripts to replicate core hand interaction functionalities, offering capabilities similar to MRTK's hand-tracking experience. Pre-Event Preparation and Support: Supported the preparation efforts for CES 2025, ensuring readiness for upcoming demonstrations and showcases.

Undertook the development and delivery of all cloud rendering project prior to June 2023, which all have been delivered to customers: • Implemented Nvidia CloudXR Client, and Paraverse Streaming Client. • Conducted application scenario tests and game compatibility tests for CloudXR Client. • Designed and implemented the phone-controller interaction solution. • Submitted the patent of depth estimation for cloud rendering stabilization. • Submitted the patent of a hand-tracking-based interaction solution. • Facilitated the development of SDK based on unique properties of cloud rendering. • Deployed the server portion of Paraverse Streaming. • Negotiated discount with AliCloud and earned XREAL an account with 40% off discount for ECS. • Invited the CEO of Paraverse to record a speech for the cloud rendering project. • Used Unity and Unreal Engine 5 to make demonstrations for cloud rendering. Participated the development and maintenance of XREAL Native SDK: • Implemented the interfaces and functions for adjusting IPD, and to acquire distortion data. • Corrected or improved the English phrasing for multiple interfaces, classes, and variables. • Implemented the code generation feature to improve development and production efficiency. This feature automatically generates the API headers files, and test source files upon compilation. • Developed the glasses server application and the native client demonstration application for internal development. • Developed the loader feature that loads the core dynamic libraries on runtime. • Developed a fundamental test application for cases requiring Android environment. The source code of the application is automatically generated upon compilation. Participate as roundtable guest for a Cloud XR Sharing Event held by a cooperating group on behalf of XREAL. (June, 2023)

Intelligent Mixed Reality v2 (Java, Python, Android Development) --Used Python to retrieve and process EXIF data from images; used Tesseract for text recognition. --Used Java to establish data interface on Android Studio. Industrial Badger App (Java, Android Development, Python, WLS) --Used Java to enable chatbot function for the application, implemented intent broadcasting, related activities and view models. Completed IBM Cloud settings to support this function. --Used Java to implement a few new features to the application, such as AnyChart, image tagging, etc. --Improved existing features, flattened the steps to start activities, and did regular bug fixing. --Implemented classes to support search functions for Azure, Box, Firebase, and SharePoint cloud services --Used Python to locally retrain an image classification model; used Java to implement a stand-alone test app and integrated it with IBA. --Used Python for the transfer learning from a ssd_mobilenet model for object detection; implemented retraining for local training and Google Colab. Used Google Cloud Vision to provide more options; integrated it with IBA

Position: Teaching Assistant for ESE590 System Methodology Jan. 2018 – May. 2018 Position: Teaching Assistant for ESE520 Agent-based Modeling and Simulation Sep. 2017 – Dec. 2017 - Lectured tutorials on AnyLogic (ESE590), and NetLogo, R, and decision theory (ESE520) - Held in-class exercises and Labs (Anylogic, Java, NetLogo, R) - Conducted exam feasibility analysis and related testing - Graded homework and exams (various topics including decision theory and system management)

Crawled millions of bid reports from websites using Python, and stored data in MySQL and MongoDB. Extracted required information such as bid winner, deal price, bid content, provider, and so forth. Built neuro-linguistic programming to improve extraction rate and accuracy, and to provide reference of prediction for future bidding events. Development of FN Note (http://www.fengniumiji.com/). Conducted a sequential event analysis for recent 10 years of data from Chinese stock market, using association analysis to show relation between certain events. Developed a web-based business intelligent platform to do the virtualization, including statistical charts and locational graphs using Baidu Virtualization API. Provided similar services to www.87.cn and guba.eastmoney.com on statistics regarding user preference, as well as lottery and stock recommendations. * Details and results of projects are confidential property of the company and are not to be disclosed.

Instruct students on their lab assignments, Q&A, grading. Evaluation of final projects.

Machine Learning; Data Mining; Professional Programming with Python and Java; Wireless Evolution Towards 5G Networks and Applications; Agent-Based Modeling and Simulation; Systems Methodology; Optimization Theory; FPGA, Software Acceleration; Digital Signal Processing; Networks and Protocols; Battery and Super-Capacitor Systems

Communications and signal processing, including image processing; embedded systems design and implementation; robotics especially robotic arms, mobile robots and autonomous quad rotors; system modeling and control; electrical and digital circuits... Undergraduate level mathematical topics, physics... General subjects in sociology, humanity, economics and management...

Communication systems, digital signal processing, and basic programming... Cultural-Topic-Based English...