Wes Ackerman Email & Phone Number

Build and refine Airflow ML training pipeline to version control training data and easily train & evaluate new models in AWS Sagemaker.Collaborate with research group to intake, test, refine, and productize new model architectures. Collaborative experiments have drastically improved f-score of vision-transformer-based image summarization model and CNN-based handwriting-recognition model on very diverse production test datasets.Analyze the output of transformer, CNN, and NLP models. Define and use metrics to identify the largest deficiencies of models, and strategize data & code refinements to improve model performance.Led a team of 3 other engineers to plan upcoming work and guide data science work on how to efficiently improve weaknesses of models. Our team produced more than 130 million ML-indexed Portuguese-language genealogical records in the last two years.

Provo, Utah Area

Designed and constructed a generative adversarial network (GAN) that generates more realistic images when performing image-to-image translation on distinct image domains.Managed deep network pipeline: image dataset creation, iteration, and normalization using PyTorch; semantic segmentation generation (using Google’s DeepLab network) and cleaning; and using scripting tools to automate data collection tasks.

Provo, Utah Area

Taught machine learning algorithms and best practices to more than 400 undergraduates and evaluated the quality of 400+ undergraduate-constructed models.Taught and debugged many types of ML model for 3+ years, providing a deep knowledge of regression models, neural networks, decision trees, clustering algorithms, KNN nearest neighbor algorithms, and ensembles (bagging, boosting).

Provo, Utah Area

Ran 100+ experiments to evaluate a stochastic multiple choice learning (sMCL) algorithm. Created a Python pipeline to process fMRI data and train sMCL CNN models.

Consulted on machine learning best practices and analyzed & explained recent scientific papers for the team.Designed methods to improve realistic image generation using laplacian pyramid structures and semantic segmentation information.

Draper, Ut

Created Java API structures to query MongoDB records and rapidly search, load, and sort hundreds of thousands of datapoints. Built front-end AngularJS functionality to display datapoints clearly based on search and sort criteria.

Madison, Wi

Designed and constructed a proof-of-concept C# application that used Epic data structures to assess the completeness of clinical documentation.

Master'S Degree, Computer Science, 3.89

Researched improving the ability of generative adversarial networks in unsupervised image-to-image translation. Have quantitative.

Bachelor Of Science (B.S.), Computer Science, 3.8

Minor in Business Management