Colloquia and Seminars

Accurate retrieval of cloud geometric and dynamical properties from satellite observations is essential for understanding atmospheric processes and improving weather and climate models. While existing stereoscopic techniques can estimate cloud-top height and large-scale atmospheric motion, recovering dense, cloud-scale vertical velocity fields at tens-of-meters spatial resolution from passive satellite imagery remains a challenging inverse problem because the required atmospheric dynamics are not directly observable.

This thesis presents a deep learning framework for estimating physically meaningful atmospheric quantities directly from passive multi-view satellite observations. A comprehensive simulation pipeline was developed by combining large-eddy simulations of shallow cumulus clouds with physically based volumetric Monte Carlo rendering, atmospheric radiometric correction, sensor noise modeling, and image alignment to generate realistic synthetic satellite imagery together with corresponding ground-truth atmospheric fields. These data were used to train a fully convolutional encoder-decoder network operating on two temporally consecutive three-view observations.

The proposed framework was evaluated on three retrieval tasks: cloud-top vertical velocity, cloud-top height, and vertical velocity at multiple fixed altitude layers throughout the atmospheric column. Experimental results demonstrate accurate retrieval of cloud-top geometry together with successful estimation of cloud-top and volumetric atmospheric motion at the native 20~m spatial resolution of the simulations. The framework recovers physically meaningful atmospheric velocity fields from passive multi-view observations and successfully infers vertical motion throughout much of the cloud-containing atmosphere using a single network architecture and observational input.

The presented results demonstrate that passive multi-view satellite imagery contains sufficient information to support dense retrieval of both geometric and dynamical cloud properties without explicit reconstruction of the three-dimensional cloud structure or intermediate physical modeling. These findings establish the feasibility of learning-based atmospheric retrieval from physically realistic simulations and provide a foundation for future development of operational data-driven cloud retrieval methods.

Udi Gal is an M.Sc. student under the supervision of Prof. Yoav Schechner.

In the Degree Realization problem with respect to a family P of graphs the input is a non-increasing sequence d = (d1, . . . , dn) of positive integers, and the goal is to decide whether there exists a simple undirected graph G ∈ P, whose degrees correspond to d, i.e., such that deg(G) = d. In this paper we consider the version of Degree Realization in which the realization is required to be a forest (i.e., P is the family for forests). We consider optimized Degree Realization in which the goal is to obtain a realization that minimizes an objective function f. That is, the goal is to find a realization G that minimizes f(G) among the realizations of the given input sequence. More specifically, we focus on the following functions: the size of an optimal vertex cover and the size of an optimal dominating set. We also consider the total and paired versions of both Min Vertex Cover and Min Dominating Set. We provide characterizations and linear time realization algorithms for all the above-mentioned problems.

Molecular dynamics simulations are computationally expensive, but optimizing them requires more than preserving exact program behavior. Many useful changes alter numerical trajectories while still preserving the physical properties that matter for a given simulation, such as energy stability, reversibility, or ensemble-level statistics.

We present a source-to-source optimization framework that searches for faster molecular dynamics implementations under physics-aware validation. The framework combines equivalence-preserving rewrites with stochastic program mutations that deliberately explore beyond ordinary semantic equivalence. For each candidate, a staged verifier checks both structural requirements and simulation-specific physical behavior, while a population-based search balances runtime, physical deviation, and program simplicity.

In this lecture, we study knapsack problems with departures under the online with a sample model. We begin with the fundamental special case of the Temp Secretary Problem with departures, where we obtain a constant competitive ratio of 1/8, providing the first performance guarantee for general instances of this problem.

We then extend our approach to the d-dimensional Online Vector Generalized Assignment Problem with Departures (VGAPWD), achieving a competitive ratio of 1/(16d) for d-dimensional resources. Lastly, we study the Multiple Knapsack Problem With Departures, which is a special case of VGAP.

For this special case, we present a more practical modification of our algorithm that achieves the same competitive ratio. Using extensive simulations on workloads derived from real cluster traces, we demonstrate that our algorithms consistently outperform state of the art algorithms and widely used heuristics, achieving typical improvements of 10–25% in total value compared to state of the art approaches.

These results demonstrate that the online with a sample paradigm successfully translates into algorithms that leverage historical data for improved empirical performance. This is aligned with the stronger theoretical guarantees we can prove within this framework.