Deep learning researcher · PhD, Technion (2026)
Models for hard-to-measure signals, with the physics and physiology built in.
My research is about deep learning for real-world temporal and spatiotemporal data: signals that are noisy, partly missing, and collected far from the lab. I bring the structure of the domain, whether physiology, physics, or network behavior, into modern models so they are not only accurate but interpretable and reliable, and I apply the same lens to large language models, asking what they actually compute rather than trusting their headline scores. I completed my PhD at the Technion in 2026 with Alex Bronstein and Reuven Cohen, and spent 2026 as an independent contractor with Apple's Video Engineering group.
Open to research scientist and applied scientist roles in LLM understanding, physics-informed AI and sensing.
Selected work
All 13 publications
Invertible Query-Key Coupling Composes with Attention Mechanisms
An identity-initialized, invertible coupling applied before scoring that adds 0.3% parameters and composes with Differential Attention, with significant gains at 150M and 455M parameters.
The Residual Stream's Effective Depth
A calibrated scalar for how much of a transformer's depth is distinct computation: fifteen of sixteen decoder-only models fall below the orthogonal-update reference, and matched controls trace the gap to correlated residual updates.
Differentiable Randers-Finsler Eikonal Solvers
A GPU fast-sweeping solver made exactly differentiable through implicit differentiation, used to recover anisotropic metrics from sparse arrival times and learn spread models on real wildfire perimeters.
From Lab to Wrist: Bridging Metabolic Monitoring and Consumer Wearables for Heart Rate and Oxygen Consumption Modeling
Full VO₂ trajectories from smartwatch and chest-strap data alone: a physiologically constrained ODE with a neural Kalman filter for heart rate, and a one-second-calibration VO₂ model at about 11% MAPE.
Research themes
More about my researchWhat transformers actually compute
Calibration-free diagnostics that audit a model instead of trusting its headline scores: how much of its depth is distinct computation, whether a personalized model conditions on the user at all, and where pre-scoring query-key geometry helps attention.
Differentiable physics and inverse problems
Classical solvers kept inside the learning loop. Implicit differentiation through fast sweeping turns anisotropic Eikonal distance fields into trainable models of wildfire spread.
Physiology from low-friction signals
Instantaneous VO₂ and cardiovascular load estimated from smartwatches, chest straps and monocular 3D pose in video, with neural ODE and Kalman structure and four datasets collected in the field.
Encrypted network traffic as time series
QUIC traces rendered as image sequences and multi-operator cellular links as multivariate series, from a 100,000-trace public dataset to real-time network selection for teleoperated vehicles.
News
All news- Completed my independent contract with Apple's Video Engineering group, where I modeled instantaneous VO₂ from 3D human pose in video.
- "Invertible Query-Key Coupling Composes with Attention Mechanisms" has been accepted to the Asian Conference on Machine Learning (ACML 2026).
- "The Residual Stream's Effective Depth" has been accepted to the Asian Conference on Machine Learning (ACML 2026).
- New paper under review: "Predicting Cardiovascular Drift from 3D Running Pose via Spatio-Temporal Graph Attention" (led by my student Yuval David).
- "Differentiable Randers-Finsler Eikonal Solvers" (presented at SGP 2026) is now published online in Computer Graphics Forum, doi 10.1111/cgf.70489.
- New paper under review: "Invertible Query-Key Coupling Composes with Attention Mechanisms".
- Submitted my PhD thesis at the Technion.
- Joined Apple's Video Engineering group as a Deep Learning Researcher, modeling physiological state from 3D human pose in video.