Autonomous AI/ML Research Scientist – Timothy K Johnsen

Resume

CV

SUMMARY

Ph.D. Candidate in Computational Science (June 2026) specializing in adaptive dynamic deep neural networks, reinforcement learning, and hierarchical frameworks for autonomous systems.

  • ML/AI Researcher with nine lead-author, peer-reviewed publications across diverse STEM domains including robotics and physical sciences.
  • Edge Computing & ML Optimization: Proven track record of engineering end-to-end pipelines to deploy resource-optimized models to constrained devices, significantly improving latency and power efficiency in noisy, real-world environments.
  • High Performance Computing with experience in supercomputer clusters, SLURM, resource and job managers, distributed and parallel computing, planetary-scale world modeling, deep learning, and multimodal large language models.
  • Interdisciplinary ML Engineer backed by science and high-performance computing experience at NASA, Lawrence Berkeley National Labs, and the University of Arizona.

SKILLS

  • Machine Learning & AI: Deep Reinforcement Learning (TD3, DQN), MLLM (Zero-Shot, gemma3), VLA (hierarchical), Planetary-Scale World Modeling, Imitation Learning.
  • Neural Network Architectures: Adaptive & Dynamic Networks, Slimmable/Gated/Split DNNs, Multi-Branch, Graph Convolutional Networks (GCNs), Multimodal U-Net, CNNs.
  • Engineering & Tools: Python, PyTorch, TensorFlow/Keras, SLURM, NVIDIA Jetson, Microsoft AirSim, Hardware-in-the-Loop (HIL), sim-to-real, Git, SQL, Cloud Computing.

EXPERIENCE

University of California, Irvine (UCI) | Intelligent & Autonomous Systems Lab

PhD Researcher | Irvine, CA | 2022 – Present

  • Architected CADENCE, an end-to-end full autonomy stack for adaptive navigation on constrained edge hardware (NVIDIA Jetson Orin Nano). Achieved a 74.8% decrease in inference latency and 75.0% reduction in energy expenditure, while simultaneously improving navigation accuracy by 7.43% over static state-of-the-art baselines.
  • Engineered hierarchical control pipelines integrating zero-shot Multimodal Large Language Models (MLLMs) for high-level task planning with dynamic 3D world building, via ray tracing from local RGB-D observations, and increasing accuracy by 27%.
  • Processed onboard sensors (IMU, GPS, RGB-D) with Deep Reinforcement Learning (TD3/DQN) for mid-level action generation and low-level actuator PID loops.

Lawrence Berkeley National Laboratory (LBNL)

Data Science / AI Researcher | Berkeley, CA | Summers 2023 – 2024

  • Developed a multimodal U-Net framework fusing over 20 years of spatio-temporal planetary data, including 10+ distinct environmental variables (radiation, NDVI), to model complex physical dynamics, achieving a high predictive accuracy (R2 score of 0.89 and RMSE of 2.36 mm/8-day) for massive-scale watershed prediction.
  • Engineered Denoising AutoEncoders with Monte Carlo sampling to repair corrupted field measurements, significantly reducing predictive bias by 4% and variance by 71% against real-world sensor artifacts (Gaussian noise and Bernoulli dropouts).
  • Managed large-scale parallel distributed training via the NERSC supercomputer to accelerate massive-scale data processing and hyperparameter optimization.

University of Arizona (UoA) | Lunar and Planetary Laboratory

Graduate Researcher | Tucson, AZ | Summer 2025

  • Executed distributed training on multimodal planetary datasets using HPC clusters.
  • Optimized transfer learning pipelines utilizing Multimodal Neural Networks (MNN) to classify geological samples, fusing dual-band Raman spectroscopy with RGB images to improve AUC scores from 0.84 to 0.90, over single-modal non-MNN approaches.

NASA Ames Research Center & SETI Institute

Research Assistant | Mountain View, CA | 2015 – 2021

  • Built a neural network from scratch in C++ to classify exoplanet spectra from geometric albedo, accelerating classification speeds and informing satellite mission planning.
  • Engineered a deployed 2D traverse system using LabView for wind tunnel surveys and a Java web-crawler to process aircraft flyover acoustic data.
  • Developed heterogeneous spectral classifiers for remote mineralogical surveys, and material classification, leading to my continued work above at UoA.

SELECTED PUBLICATIONS

  • Johnsen, T. K., et al. “CADENCE: Context-Adaptive Depth Estimation for Navigation and Computational Efficiency.” IEEE World AI IoT Congress (AIIoT), 2026.
  • Johnsen, T. K., et al. “Denoising autoencoder for reconstructing sensor observation data and predicting evapotranspiration…” Water Resources Research, 2025.
  • Johnsen, T. K., et al. “NaviSplit: Dynamic multi-branch split dnns for efficient distributed autonomous navigation.” WoWMoM. IEEE, 2024.
  • See a full list at: https://scholar.google.com/citations?user=nvidNbAAAAAJ&hl=en

Awards: Grad Slam Semi-Finalist (UCI 2026, SJSU 2021), ACSESS Poster Award (SDSU, 2024 & 2025), ASSICS Fellowship (2021-2023).

Presentations: Featured speaker/presenter at American Geophysical Union (Planetary Modeling with ML), the Life and Space Conference (AI for Planetary Exploration), IEEE (Drones/AirSim for Wildfire Surveillance), NASA Ames Research Center, and UC Irvine.