Caleb J Tolman Email & Phone Number

Atlanta, Ga, Us

Marietta, Ga, Us

San Francisco, California, Us

Berlin, Be, De

A few small freelance consulting projects between full-time employment. Mostly advised about building feature-store ML pipelines, data augmentation, and bandit algorithms

서울특별시, Kr

Palo Alto, Ca, Us

Roadmap planning lead, project milestone advisor, incident manager, employee mentor. In charge of all DS, Eng, and PM ICs. Position was created to mirror the DSP VP's role but from a technical perspective. Lead a "turnaround quarter" where we did a major code refactoring project and built a DS flywheel. Revenue doubled after fixing bugs the first quarter, and up another 40% the next. We were then acquired by Applovin

Palo Alto, Ca, Us

Technical leader in the mobile marketing department, advancing industry-innovative end-to-end solutions via coordination with Eng, Product teams, and management, often at enormous scale.

Palo Alto, Ca, Us

Due to the success of its games and other platforms, MZ makes more marketing decisions and creates larger datasets than perhaps any of its peers. Our team is building machine learning and optimal exploration algorithms on these datasets, using Hadoop, Java, PostgreSQL, HDFS, Spark, Scala, Python, NumPy, Shell, Unix, Tensorflow, R, and more. We strive for elegant solutions based upon the most advanced mathematics. Instead of blindly accepting every mathematical assumption common to the industry, we often challenge these assumptions, test them, and exploit the degree to which they do not hold. We also use our analysis to request major changes in the industry as a whole, and view our company as a leader in the development of mobile marketing.

This is a hobby that grew into a long-term, part-time personal business---Private swimming instruction for all levels and ages in Austin Texas, and then Palo Alto. Hundreds of students over 10 summers (Over 16 years, skipped 6). Mostly beginners, mostly ages 5-12, yet several adults and several that advanced to be competitive. I loved this job, and miss all my students.

Redwood City, Us

Liftoff is a high-performance mobile marketing broker (DSP), for those wanting not just installs but engaged and valuable app users. They have perhaps the highest quality engineering and customer service in the industry. It was my task to "beef up" their machine learning after their series A. This ranged from the simple, such as: proposing, designing, building, and tuning their Age and Gender models---intermediate predictions that improve later predictions and also allow more detailed reporting for clients. It also included the more advanced, such as: look-alike-models that solved the cold-start performance problem for new clients, adjusting models for false negatives caused by the delay in attribution, and several applications of dimensionality reduction for adding third-party and sparse data.

New York, Ny, Us

Using Chapter 11 Filings to determine the likelihood that our clients will successfully emerge from bankruptcy. (Multi-factor Dimensionality Reduction). Started with a separating hyperplane model, but then it evolved over time to be a boosted-tree model with small economic models built in as base features. This was a novel approach to apply transfer learning: we could use complex decision models that looked at company financials to determine the "value at risk" of making too many motions. If a company had nothing to risk, they often would file every possible motion in court. These value-at-risk scores perfectly addressed that issue, leading to much improved prediction power, even on small datasets

Us

Managed and optimized all internet advertising accounts to generate affordable and quality traffic for our Clicksmart.com online yellow pages. I also advised the CEO and Executive Board on four of their most crucial business decisions during my short stay: 1) Talent Attribution Models---when some referral-lists are known to be much better than others, and most salesmen are assigned only a few lists, correctly discern the relative skill of each salesman and design a fair compensation plan. Latent Factorization. 2) Churn models. 3) Estimating price elasticities--concluded that we did not have enough data, and thus should outsource to a broker who did. 4) Retention models--an extension of the churn models that identified which customers could be profitably retained or retargeted.

Austin, Tx, Us

Infant (6-18mo) Cognition Study. Assisted in the implementation of two studies to determine: 1) At which age memory chunking begins, & 2) At which age cause-and-effect understanding begins.

Ph.D., Management Science & Engineering

M.A., Management Science & Engineering

B.S., Mathematics