- Leading the development of the framework for in-silico generation of multi-functional proteins using diffusion models- Developed a high throughput pipeline for motif-scaffolding and validation of generated proteins, capable of generating more than 4k proteins per hour- Led the campaign for the production of over 2M multi-functional proteins for drug delivery

Continuous Spatiotemporal TransformerModeling spatiotemporal dynamical systems is a fundamental challenge in machine learning. Transformer models have been very successful in NLP and computer vision, but they are discrete models and have no guarantees regarding continuous sampling. We present the Continuous Spatiotemporal Transformer (CST), a new transformer architecture that is designed for modeling of continuous systems. This new framework guarantees a continuous and smooth output via optimization in Sobolev space. We benchmark CST against spatiotemporal dynamics modeling methods and achieve superior performance in a number of tasks on synthetic and real systems.Neural Integral EquationsIntegral equations (IEs) are equations that model spatiotemporal systems with non-local interactions. While efficient algorithms exist for solving given IEs, none can learn an IE and its associated dynamics from data alone. We introduce Attentional Neural Integral Equations (ANIE), where the integral is replaced by self-attention, which improves scalability and model capacity. We demonstrate that (A)NIE outperforms other methods in both speed and accuracy on several benchmark tasks in ODE, PDE, and IE systems of synthetic and real-world data.

Neural Integro-Differential Equations Modeling continuous dynamical systems from discretely sampled observations is a fundamental problem in data science. We introduce the Neural IDE (NIDE), a framework that models ordinary and integral components of IDEs using neural networks. We test NIDE on several toy and brain activity datasets and demonstrate that NIDE outperforms other models, including NODE. We show that NIDE can decompose dynamics into its Markovian and non-Markovian constituents, via the learned integral operator, which we test on fMRI brain activity recordings of people on ketamine. Altogether, NIDE is a novel approach that enables modeling of complex non-local dynamics with neural networks.

Hardware and software development for behavioral studies. Main projects under development:1) VocalMat (https://elifesciences.org/articles/59161) It is known that ultrasound vocalizations (USVs) are a crucial source of information for behavioral studies in several different species. VocalMat brings an automated algorithm that is able to remove noise from recorded audio files and extracts characteristics from each vocalization recorded. This tool is getting new features in order to correctly assign each vocalization detected to a single individual during a social experiment. With such knowledge about how the animals vocalize under certain conditions, several new questions can be asked. 2) MamaBot There are many unanswered questions about the relationship between a mother mouse and the offspring regarding the impacts of lacking maternal care during the first stages of life and what the real needs of a newborn mouse are. MamaBot is a portable feeding system driven by the way how newborns interact with an artificial nipple and have the potential to answer many questions about newborn’s behavior. 3) idTracker + RFID Owing to its high temporal and spatial resolution, video tracking is the main method used in the laboratory to track animals in a group in order to study their behavior. Once marking the animals can potentially modify their behavior, correctly identifying the subjects after a crossing is really challenging. idTracker brings a multi-tracking algorithm that extracts a characteristic fingerprint from each animal in a video recording of a group. It then uses these fingerprints to identify every individual throughout the video and uses information gotten by the RFID anthems to perform necessary corrections.

Working on:- Reliability engineering;- Software Quality Assurance (SQA);- Systems automation.

- Microsoft Visual Studio;- C/C++- Development of software to controlling the actuator of a driller robot arm based on image processing;- Software development to "stitch" images from several camera in order to produce single panoramic image with 360-degree view.

Design and implementation of distance measurement algorithm in Matlab/Simulink and visual tracking of a moving object in unstructured environment using single camera.Knowledge acquired: - LMS Virtual Lab experience;- SolidWorks;- Image processing (target identification in images without stable background) using Simulink tools.

Dozens Simultaneous Tracking of Objects.Knowledge acquired: - C/CUDA programming language;- Parallel processing theory;- Image processing (target identification in images with noise) ;- Monte Carlo, Kalman and Particule Filters applied to multiple target tracking.

Quantification Visual Automatic Response freezing in mice.Knowledge acquired: - Matlab programming language;- Image processing (edge identification and movement quantification in images);- Machine interface development (theory and implementation in Matlab).