Skip to main content

Curriculum Vitæ

📞 +1 415 779 4282📨 hello@🌐 https://leif.sh/cv/

Leif is a computational modeling expert with a background in machine learning, speech recognition, and human visuomotor behavior. His industry experience has focused on computational modeling in areas spanning robotics, human-computer interaction, speech recognition, and translation. Leif's work in academia focused on computational models and machine learning applied to human visual and motor behaviors.

Industry Experience

Waymo

Safety ResearcherAug 2021 – presentRemote / SF Bay Area

Reference Behavior Modeling
    kinematicsstatistics
Developed safety envelope models for evaluating autonomous vehicle behavior.

Google

Software EngineerFeb 2016 – Aug 2021Mountain View, CA

Sequence Models for Attention
    convolutional-neural-networkflumetensorflow
Developed and evaluated sequence-based gaze models, as well as downstream filtering for existing single-frame models.
Federated Speech RNN-Transducer
    federated-learningtensorflow
Scaled on-device training to a full-size RNN-T model with ~10⁸ parameters; previous on-device models had ~10⁵ parameters.
Gboard Spatial Modeling
    c++statisticstensorflow
Developed models for personalized spatial inference, and supervised another engineer during implementation and testing.
Gboard User Study
    eye-trackingmediapipestatisticsuser-study
Ran and analyzed a small study of keyboard users that included a head-mounted eye tracker, using insights to inform product direction.
Gboard Federated Learning
    federated-learningtensorflow
Built infrastructure for training federated models, enabling modeling teams to train on-device or server-side without code changes.
Speech G2P RNN-Transducer
    c++decodingtensorflow
Improved an existing Grapheme-to-Phoneme model for US English using an RNN-Transducer.

Research InternSummer 2009 & 2010Mountain View, CA

Speech Discriminative Language Model
    map-reducevoted-perceptron
Used a distributed discriminative language model to improve speech recognition performance for weak acoustic models.
SThedinoateabean
Translation Parsed Language Model
    dependency-parsingmap-reduce
Used a dependency parser to augment language models for machine translation.

Software EngineerNov 2004 – Mar 2008Mountain View, CA

    bigtablec++mysqlperforcepython
Scaled a customer support tool from handling xxxk messages in MySQL to xxxM messages in Bigtable.

Sutros

Software EngineerMar 2008 – Aug 2008San Francisco, CA

    aws-s3javascriptmysqlrails
Developed a web tool to help musicians license their music under the Creative Commons.

Salomon Automation

Research InternAug 2002 – May 2003Graz, Austria

    pythonjavaantlr
Worked with staff researchers to implement a set of CORBA component generator tools.

Academic Experience

The University of Texas at Austin

PhD, Computer ScienceAug 2008 – Dec 2015Austin, TX, USA

My graduate studies were supervised by Dana Ballard in the Vision, Cognition, and Action Virtual Reality Lab. My dissertation focused on the use of efficient codes in complex feedback processes like movement and speech. I also gained proficiency working with several unusual technologies, including eye trackers, motion capture, and virtual reality.

Teaching Assistant Experience

Coursework

North Carolina State University

BS with honors, Computer Science
BA with honors, Multidisciplinary Studies
BS, Applied MathematicsAug 1997 – May 2002Raleigh, NC, USA; Grenoble, France

North Carolina School of Science & Mathematics

DiplomaAug 1995 – May 1997Durham, NC, USA

Publications🎓

📝L Liu, L Johnson, O Zohar, DH Ballard2019Humans Use Similar Posture Sequences in a Whole-Body Tracing TaskIscience 19, 860-871

📝KC Sim, F Beaufays, A Benard, D Guliani, A Kabel, N Khare, T Lucassen, P Zadrazil, H Zhang, L Johnson, G Motta, L Zhou2019Personalization of end-to-end speech recognition on mobile devices for named entitiesProc. IEEE Conf. on Automatic Speech Recognition and Understanding (ASRU)

📝R Prabhavalkar, K Rao, T Sainath, B Li, L Johnson, N Jaitly2017A Comparison of Sequence-to-Sequence Models for Speech RecognitionProc. Interspeech

📝A Bruguier, D Gnanapragasam, F Beaufays, K Rao, L Johnson2017A more general method for pronunciation learningProc. Interspeech

I Karpov, L Johnson, R Miikkulainen2015Evaluating team behaviors constructed with human-guided machine learningProc. IEEE Conf. on Computational Intelligence in Games (CIG)

M Tong, S Zhang, L Johnson, DH Ballard, MM Hayhoe2015Modeling Task Control of GazeJournal of Vision 15(12):784

📝L Johnson, B Sullivan, MM Hayhoe, DH Ballard2014Predicting human visuomotor behaviour in a driving taskPhil. Trans. R. Soc. B 369

L Johnson, DH Ballard2014Efficient codes for inverse dynamics during walkingProc. Assoc. Advancement of Artificial Intelligence (AAAI)

L Johnson, DH Ballard2014Classifying movements using efficient kinematic codesProc. Ann. Mtg. of the Cognitive Science Society

L Johnson, J Cooper, DH Ballard2013Unified loss functions for multi-modal pose regressionProc. IEEE Intl. Joint Conf. on Neural Networks (IJCAI)

L Johnson, B Sullivan, DH Ballard, MM Hayhoe2013A soft barrier model for human behavior in a two-task driving environmentProc. Ann. Mtg. of the Cognitive Science Society

📝B Sullivan, L Johnson, C Rothkopf, MM Hayhoe, DH Ballard2012The role of uncertainty and reward on fixation behavior in a driving taskJournal of Vision 12(9)

P Jyothi, L Johnson, C Chelba, B Strope2012Distributed discriminative language models for Google voice-searchProc. IEEE Intl. Conf. on Acoustics, Speech and Signal Processing (ICASSP)

P Jyothi, L Johnson, C Chelba, B Strope2012Large-scale discriminative language model reranking for voice-searchProc. N. Am. Assoc. for Computational Linguistics - Human Language Technologies (ACL-HLT)

E Teiniker, S Mitterdorfer, L Johnson, C Kreiner, Z Kovacs, R Weiss2003A Test-Driven Component Development Framework based on the CORBA Component ModelProc. 27th Ann. Intl. Computer Software and Applications Conf

Posters

L Johnson, DH Ballard2015Modular Motor Control in Full-Body, Goal-Directed MovementsAnn. Mtg. of the Soc. for the Neural Control of Movement (NCM)

L Johnson, DH Ballard2014Analysis of Human Movement During a Virtual Tracing TaskBiann. Mtg. for Natural Environments, Tasks, and Intelligence (NETI)

Other

🎬L Johnson2013Python tools for coding and feature learningSciPy

Miikkulainen, E Feasley, L Johnson, I Karpov, P Rajagopalan, A Rawal, W Tansey2012Multiagent Learning through NeuroevolutionJ. Liu et al., eds., Advances in Computational Intelligence, LNCS 7311, 24-46, Berlin, Heidelberg: Springer

B Sullivan, L Johnson, DH Ballard, MM Hayhoe2011A modular reinforcement learning model for human visuomotor behavior in a driving taskProc. Active Vision Symposium, Artificial Intelligence and the Study of Behavior

L Johnson, P Wurman2002Information and Product Quality Dynamics in Tiered Supply NetworksProc. AAAI Workshop on Multi-Agent Modeling and Simulation of Economic Systems

Other

Technology

Hobbies & Interests

Languages