Sergei V. Kalinin

66 Followers
122 Following
22 Posts
Sergei Kalinin on LinkedIn: Post-Experiment Forensics and Human-in-the-Loop Interventions in…

Explainable automated experiment and human in the loop interventions Development of automated experiments (AE) in microscopy and other physical imaging and…

Sergei Kalinin on LinkedIn: Atomic-scale electrochemistry on the surface of a manganite by scanning…

Disruptive ideas in science: there is no (pure) physics of oxides on the nanoscale One of the most interesting processes in scientific community is the…

Sergei Kalinin on LinkedIn: High Output Management

Management and management culture (for academic scientists)* One of the main learning experiences I got over the last year at #amazon has been the management…

Sergei Kalinin on LinkedIn: Exploring physics of ferroelectric domain walls via Bayesian analysis of…

Instrumental communities, and what makes a breaking point One of the greatest books on the history of science I have read is "Instrumental community" by Cyrus…

Sergei Kalinin on LinkedIn: #ml #autonomous #materials #microscopy #bayesian #quantumtechnology

What will 2023 bring for ML (in experimental physical sciences)? ML has been the most dynamic fields for the last decade, so there are enough predictions for…

It is crucial to set terminology in a new field early on. Currently with the forensics for automated experiments, I realize that we already have "regret of a mature model" as a part of emerging definitions.... I guess because it has a memory, unlike contextual bandit:)
Sergei Kalinin on LinkedIn: My journey into machine learning started about 15 years ago, as a part of…

My journey into machine learning started about 15 years ago, as a part of development of spectroscopic modes for Scanning Probe Microscopy. At that time…

Atomic-scale e-beam sculptor

A system and method (referred to as the system) fabricates controllable atomic assemblies in two and three dimensions. The systems identify by a non-invasive imager, a local atomic structure, distribution of vacancies, and dopant atoms and modify, by a microscopic modifier, the local atomic structure, via electron beam irradiation ...

The interesting problem for workflow hyperlanguages will be translation from the lab with one set of instructions to another. On the positive side, once expressed in the same units, the economics of orchestration follows immediately.
The necessary addition to data and code sharing is the physical workflow sharing. Which of course requires hyperlanguage to write them, and testing and calibration procedures to make them transferable between labs. Imagine the ML codes have to run on GPUs with different precision