Ben Murdoch Email & Phone Number

Mountain View, Ca, Us

Currently working on agents for multi-hop question answeringContributed to 3rd party code generation models Worked on code question answering for Google Search (SGE)Led our safety implementation and responsible AI efforts for the launch of CodeTips, Colab, and Android Studio

Sunnyvale, California, Us

Built natural language classifiers for social engineering attacksCreated an auto-nlp framework with Huggingface, Kubernetes and Ray-Tune Ran our deep learning mentorship programSpoke at RSAC on generative email fraud

Salt Lake City, Utah, Us

Managed the world's largest production deployment of GPT2 (600M+ model calls per month)Doubled the speed of model inference and reduced model serving costs by 65%Built my own community driven version of MTurk and collected tens of thousands of labelsImplemented and auto ML pipeline to greatly reduce the complexity of T5 model trainingTrained peers to build their own custom deep learning models to solve a wide range of tasksSetup & managed external research partnershipsLead ML reading groups, hackathons, and community events. Try it out at: play.aidungeon.io

Sunnyvale, Ca, Us

Fine tuned BERT for domain specific conversational language modelingWorked on the Alexa AI Conversational Modeling team

Redmond, Washington, Us

Worked with Scientists from MSAI & Microsoft Research to design and implement novel methods for personalized search in Microsoft TeamsBuilt an encoder decoder pipeline for free-form named entity generation from documents

Provo, Ut, Us

Worked on reinforcement learning for text based adventure games, natural language generation, dataset collection, and emergent communication in order to facilitate human robot interaction. Some of my work includes:IJCAI 2017: Affordance Extraction via Word Embeddings https://www.ijcai.org/proceedings/2017/144Our work was featured across many news outlets including TechCrunch and ScienceDailyWorkshops:NIPS - ViGIL 2017: Informing Action Primitives Through Free-Form Text AAAI - KEG 2018: A Semantically-Informed Algorithm for Extracting Interaction ModesOther News about my Research:news.byu.edu/news/researchers-use-wikipedia-give-ai-context-cluespcc.cs.byu.edu/category/eve/

Redmond, Washington, Us

Helped build the first scalable deep learning architectures for domain specific named entity recognition and slot tagging in Cortana. I worked directly on production optimization of recurrent and convolutional neural networks.

Pleasant Grove, Utah, Us

Directed leadership training exercises for a diverse range of high school and college studentsManaged and trained new volunteers and staff how to run and operate our custom simulators