Jacob H. Rothschild

GPA: 5.7/6.0

Thesis: Solving Min-Cost Concave Generalized Dynamic Flows and Approximating Dynamic Optimal Power Flows

• Awarded a NOK 150 000 (approx. $16k) scholarship from Jansons Legat.
• Awarded other scholarships totaling SEK 137 000 (approx. $14k) beyond public funding.
• Teaching assistant in the course Network & Integer Optimization.

Major: Mathematical Statistics

Thesis: Deep reinforcement learning for real-time power grid topology optimization

• Achieved the maximum grade in 19 out of 20 courses.
• Awarded one-year, tuition-waived, bachelor’s level exchange studies to UC Berkeley (later cancelled due to the pandemic), as the only student from the Faculty of Science.
• Awarded scholarships totaling SEK 20 000 (approx. $2k) beyond public funding.

GPA: 5.6/6.0

• Placed 9^th nationally in Norsk Informatikkolympiade 2017 (nationwide informatics competition for Norwegian high school students) out of 245 participants.

• Built systems for autonomous and optimal business decisions by combining causal inference, survival analysis, and machine learning.
• Led the design of a randomized controlled trial spanning €2B+ in insurance claims.

• Analyzed causalities from driving surface of dumper trucks to accelerometer data.
• Created and implemented algorithm for detecting loading times of dumper trucks from accelerometer data.

• Created machine learning (ML) benchmarker and metric for document classifier.
• Improved document classifier by up to 35%, using semi-supervised ML.
• Created critical sync service from Microsoft SharePoint to Google Cloud Storage.

• Built Spark architecture on Dataproc, for distributed ML, simultaneously achieving a 300x speedup and an 80% cost reduction.

• Created an official example for the Fusetools platform.
• Improved the cross-platform Kudos (now Spotlite) app.
• Had my story covered in Norway’s largest newspaper (Aftenposten).

• Created the iOS app SKAMvarsler, which became the #1 paid app in the Norwegian App Store and was covered by Norway’s second largest newspaper (VG).
• Created the Android game Defend n’ Send, which today has 5,000+ downloads.
• Created the native Android and iOS app AVM-tool for Vibratec Akustikprodukter.
• Created the native Android and iOS app Q-system for Norwegian Geotechnical Institute, which today has 1,000+ downloads and a rating of 4.6/5.0.

Reimplemented ACT and Diffusion Policy; beat every public vanilla-ACT result on PushT (+14pp); wrote explainers (350K+ views). Writeup · Robot Control · ACT

Created website finding cheapest way to fulfill whole shopping carts. Involved MILP, web scraping, data pipelining, resource management, frontend. Website

Worked through all of Sutton & Barto’s reinforcement learning book (including exercises) and implemented most of the algorithms therein. GitHub

Studied and built GPT-2 from scratch on the level of the nn.Parameter primitive in PyTorch, then trained it on a multi-GPU setup. GitHub

Studied and implemented DeepPoly certification for ANNs, CNNs and ResNets on the level of PyTorch tensors. GitHub

Developed a fraud detection system to help the NGO IMPACT Initiatives in their mission to improve the impact of humanitarian aid programs.

Organized 10 days of introduction activities for up to 60 international students.

Volunteered in Cartagena, Colombia for six weeks, teaching English in the slums and dance in a drug rehabilitation center for teenagers.

Two days on AI and entrepreneurship. 20% acceptance rate.

One-week hackathon on the topic of urban futures.

11-week course. Obtained perfect grades.

One-year program for self-development and civic engagement. 17% acceptance rate.

One-year program for self-development and leadership. 34% acceptance rate.