Internship Opportunities

Traditional Structure from Motion (SfM) and Simultaneous Localization and Mapping (SLAM) systems typically output sparse 3D point clouds that lack explicit information about the underlying surface geometry. This research project focuses on exploiting local affine transformations between image pairs to extract significantly richer geometric data. By mathematically analyzing how local image patches deform between views, based on known or estimated affine transformations, it is possible to compute not only the precise 3D coordinates of a point but also its corresponding surface normal. This novel approach directly yields the orientation of the observed surfaces, seamlessly bridging the gap between sparse point extraction and dense surface reconstruction.

As an integral part of an ongoing, specialized research initiative, this project offers students the opportunity to dive deep into projective geometry and advanced computer vision algorithms. The primary objective is to implement, refine, and test these affine-based mathematical models to reliably estimate normal vectors within standard SfM/SLAM frameworks. Students will be tasked with validating the methodology using diverse image pairs, analyzing the accuracy of the computed surface normals, and exploring how this enriched spatial data can improve subsequent tasks such as dense meshing or robust camera tracking. This is a unique opportunity to contribute directly to cutting-edge research and expand the theoretical boundaries of 3D vision.

Literature Review and Problem Analysis: Investigate existing AI-driven compiler optimization techniques and their application to HPC codes. Analyze the unique characteristics of stencil computations on structured meshes and identify optimization opportunities.
Using AI Models for Compiler Optimization: Experimenting with LLM models to predict optimal compiler transformations (e.g., loop tiling, fusion). Incorporate domain-specific features such as data locality, memory access patterns, and computational dependencies.
Implementation and Integration: Integrate the developed AI models into an existing compiler framework (e.g., LLVM). Implement a feedback-directed optimization loop to iteratively refine performance based on runtime profiling.
Performance Evaluation: Benchmark the optimized stencil codes across various HPC architectures (e.g., CPUs, GPUs). Compare performance improvements against traditional compiler optimizations in terms of execution time, energy efficiency, and scalability.

The task is to develop a simulator environment to support algorithmic and hardware development, based on an existing simulator system. Improvements to parts of the current system are necessary, such as the rendering engine (Ogre vs. FlightGear, testing with new versions) and the testing and communication with image processing hardware (GPU, FPGA). A very important part of the work is also to write comprehensive documentation on the setup.

Keywords: Unmanned Aerial Vehicles (UAV), FPGA, GPU, collision avoidance, video and image processing

Neurons of the nervous system display considerable variability in terms of their shape (morphology), spontaneous and input-driven physiological behavior, and molecular composition (markers). Based on these characteristics, neurons are grouped into cell types, which are thought to have distinct functions. Experiments carried out at the Institute of Experimental Medicine have yielded a large database of the anatomical and physiological features of hippocampal cells. Using this database, we aim to determine the number of different cell types in the mouse hippocampus, and to identify the relations between different types of neuronal characteristics, through the application of methods in modern statistics and machine learning. We will also characterize the multi-dimensional variability of cell properties within individual cell classes, and construct populations of model neurons that match real neuronal populations not only in terms of their average properties, but also regarding their variability. In the long run, these studies will allow us to understand how the diversity of neurons contributes to the efficiency and robustness of computations in the brain, and may facilitate the construction of artificial systems with similarly robust and efficient performance.

Some programming experience is required; knowledge of modern statistics and machine learning methods would be an advantage.

The most important requirement is a strong interest in neuroscience. Some programming experience and a working knowledge of English are also needed.

The pharmaceutical industry relies heavily on solid dosage forms, and the tablet forms the basis of many preparations, including innovative ones (traditional, compression-coated, multilayer tablet systems, compression products containing liquid elements, effervescent tablets, and orally disintegrating tablets, or ODTs). The use of compression simulators and the analysis of compression force curves are essential for tablet design. The elastic, plastic, and fragmentation behavior of materials can be characterized by analyzing these curves. The thesis aims at the mechanical and IT background of this. The candidate performs the laboratory tasks using the Diaf eccentric tableting machine and the Natoli NP-RD10A compression simulator located at the Faculty of Pharmacy of Semmelweis University. The aim of the practical work is to equip an eccentric tablet press, used in compression tasks, and to write a related program, with the help of which tablets with precisely designable parameters can be compressed, knowing the compression force curves. We are looking for applications from students who are receptive to mechanical engineering tasks.

The ocular drug penetration through biological barriers is a little-studied area. The current ophthalmic formulations are not effective enough due to the low bioavailability (~5%) and insufficient tissue penetration. The eye drops should be applied frequently (5-6 times daily), which is not convenient for the patients. Also the current therapy for retinal disorders (like macular degeneration) is invasive, because intravitreal injections should be applied repeatedly to treat the posterior part of the eyeball of the patients. In the laboratory of Barrier Modelling and Innovative Pharmaceutics new, innovative ocular formulations are under development. To evaluate them special microfluidic chambers should also be developed. The task is to test the novel formulation on corneal and retinal tissues incorporated into the microfluidic chambers.

The widespread use of unmanned aerial vehicles (UAV) increases the risk of accidents; therefore, new procedures must be developed to avoid dangerous approaches and collisions. Due to the size of small UAVs, only passive image processing systems can be used. However, high-resolution cameras are required to detect dangerous aircraft from a large distance. High computing performance is required to process real-time image streams arriving from the cameras, which can only be achieved quickly and efficiently by using application-specific image processing architectures. One possible way is to use an FPGA, which allows high computational power with relatively low power consumption and reasonable development times. The task is to develop an FPGA system that runs the previously developed and tested SAA algorithm.

Colony picking is a frequent task in microbiology, where individual colonies are sampled and transferred into a liquid medium for further culturing. In this project, we will be developing a colony-picking robot to automate this otherwise tedious task. The goal is to develop a system that can be either used directly by the user or integrated into a laboratory automation platform for high-throughput sampling. The student will take part in the development of a hardware/software prototype, depending on her/his interests.

In modern microscopy systems, motorized microscope stages are crucial. Without motorized stages, imaging a large number of samples in high-throughput experiments or imaging large areas is unimaginable. Unfortunately, commercially available motorized stages have a prohibitive price, and their precision usually far exceeds what is generally needed. Luckily, with the spread of hobby 3D printers, laser engravers, and CNC machines, mechanical and electrical components (stepper motors and drivers, rail systems, microcontrollers) are now readily available. Although there are several open-source stages in the literature, their integration into image acquisition platforms is still problematic [1-2]. The task of the prospective student will be to take part in the development of a prototype and/or to write software to control the microscope stage. 

Students can focus on a subset of the project depending on their interests and available time:

Literature review: Review current literature that is relevant to the design of the hardware/software components.
Choose microscope platform: Prospective students can choose to motorize our manual microscopes.
Design: Design hardware components: Design a custom stage for the microscope of choice. This will encompass both 3D mechanical design and the design of the electronics (Arduino, stepper motors, power supply, etc.).
Prototyping: Using the university fabrication facility, build a working prototype.
Software development: Create a simple application to control the available hardware.

One major therapeutic limitation of currently available small-molecule drugs is their limited action specificity. Limited specificity can manifest as serious side effects and as a loss of therapeutic potential, especially for diseases affecting the nervous system. Gene therapy methods have the potential to target specifically those cell types that cause a particular disease, without altering the function of the rest of the nervous system. In this project, we will use experimental animal models to test new viral-based gene therapy approaches by analyzing gene expression specificity across brain areas and cell types. The candidate will be mostly involved in tissue processing, microscopy, and image processing tasks.

Generative networks have recently undergone significant development and are now able to produce text and images with a high degree of fidelity at large scale. The aim of this topic is to investigate the dynamic properties of these models when considered as autonomous systems. In other words, to what extent do they generate new content starting from a given prompt, or converge to constant, periodic, or even chaotic dynamics after a certain time?

Given a 3D Gaussian Splatting (GS) model of an environment reconstructed from RGB video, parts of the environment may have changed since the original recording. The aim of this project is to develop an efficient incremental update method for the GS model, so that local environmental changes do not require rebuilding the complete 3D model.

The goal of the projects is to develop Maxwell equation solvers and graphical user interfaces specifically designed for metamaterial research. A set of freely available parallel Finite Difference Time Domain, Finite Element, and Method of Moments codes will be developed to simulate the propagation of electromagnetic waves in metamaterials and photonic crystals. The algorithms will be supplemented with homogenization procedures for metamaterials and composites.

Electromagnetic field computation software development for metamaterial and photonic crystal design, which can be run on a supercomputer.

Combining different investigation methods of cell-cell interactions, we aim to understand the behavior of multicomponent microbial populations. Spreading or extinction of individual strains will be studied with respect to local interactions, initial cell concentration, and spatial structure. In this interdisciplinary project, bioinformatics, comparative genomics, and microbial experiments are combined to investigate how the cell’s environment and its interactions control the behavior of the individual and influence the population. Numerous yeast strains will be investigated in single-strain and pairwise co-cultures in different environments (by variation of carbon source and density) to identify fitter strains. Based on these experimental results, mathematical models will be built to simulate interactions and predict multicomponent colonies that can be tested experimentally.

The role of yeast in understanding molecular processes has already been remarkable. In our project, we would like to gain a clearer picture of colony formation and cellular interactions. Wet lab opportunities include following the growth of yeast cells, modifying their genome, and observing the formation of different colony structures. Based on these results, a model has been developed to simulate cellular interactions and the predicted behavior of multicomponent colonies tested in the lab. Students are offered the opportunity to take part in all parts of the project, from the planning through the implementation to the evaluation of the results. Individual ideas in connection with the project are also welcome and can be carried out if feasible.

The implementation of face recognition and tracking technology on UAV swarms is a promising development direction that has the potential to enhance security, improve surveillance, and aid in search and rescue operations. The ability to identify and monitor individuals from the air is a powerful tool that can help create a safer environment in a large work area.

During the last decade, a novel interdisciplinary scientific area, the research of metamaterials, emerged. The fundamental idea behind metamaterial research is that the interaction of electromagnetic waves with matter is governed not only by chemical composition, but also by structuring, which can produce novel properties that are not available in bulk form. The research field of metamaterials extends the limits of subwavelength imaging, makes it possible to cloak scattering objects, can enhance the performance of antennas, and can increase the sensitivity of sensors. Nowadays, the research of metamaterials has advanced to a level where commercial applications are underway.

This thesis aims to revisit the challenge of gene prediction by using advances in transformer-based neural networks and genomic language models. With the increasing availability of genomic data, a vast portion remains unexplored, leaving numerous genes and proteins unidentified. This research intends to apply cutting-edge genomic language models, trained on extensive sequence datasets, to address the limitations of current gene prediction methods. The project will focus on developing a more robust, generalizable tool capable of uncovering new insights into the genetic fabric of uncharted genomes. Students will explore the application of these advanced AI techniques to improve the accuracy and efficiency of gene prediction, potentially improving our understanding and identification of novel genetic elements across diverse species. The ultimate goal is to contribute to the broader bioinformatics field by providing a sophisticated approach to predict and annotate genes in newly sequenced genomes, thereby facilitating further biological discoveries and advancements. Keywords: molecular biology; deep learning; genomic language models; PyTorch; HPC

Understanding and reviewing the relevant literature; development of gene prediction algorithms based on genome sequence representations

The neuronal response to the repeated presentation of identical stimuli is never exactly the same; in fact, it is highly variable, while perception is remarkably stable. In this project, we are interested in characterizing this variability, including its possible benefits and limitations for learning, memory, and navigation. Specifically, during navigation in virtual reality, the hippocampal code for space changes from day to day, a phenomenon called representational drift. Our goal is to study the hypothesis that representational drift is not completely random but avoids certain directions in the state space. This would enable the system to quickly learn novel information while also maintaining a stable representation of earlier memories.

Review the recent papers analyzing representational drift in the hippocampus and related cortical areas, with special emphasis on the methods used to characterize the directionality of the drift. Develop novel analysis methods to detect systematic components of the drift, and test these methods on synthetic data. Finally, apply the analysis techniques to real data recorded in mice navigating in virtual reality environments.

The scope of the project is technology-agnostic, allowing for the exploration of various 3D reconstruction methodologies. This could range from identifying bottlenecks and extending established photogrammetry pipelines, such as accelerating feature extraction, matching, or bundle adjustment in COLMAP, to optimizing newer neural, volumetric, or SLAM-based approaches. The primary task is to apply targeted hardware acceleration techniques and benchmark the resulting performance against standard implementations, ultimately demonstrating a clear pathway toward robust, real-time mapping solutions suitable for autonomous navigation, robotics, or augmented reality.

For the separation of the epidermis (consisting of the stratum corneum and viable epidermis) and dermis, mechanical, thermal, and chemical techniques are available. Dermatomed skin can be prepared by means of an electrical dermatome, which performs surface-parallel sections composed of the epidermis, including the stratum corneum, as well as parts of the dermis. Normally, a slice thickness of 200–400/500 μm is recommended for human skin (OECD 2004b; USEPA 2004). In doing so, the hair follicles will be cut, but the resulting holes will rapidly close during incubation with the acceptor medium due to swelling of the tissue, especially if aqueous donor and receptor fluids are used. During dermatomization, care must be taken to avoid mechanical damage to the SC, the main barrier of the skin. To overcome this problem, protection of the skin surface by a plastic foil is indicated (Brain et al. 1998). Typically, dermatomed skin is used for permeation studies, especially if the influence of the hydrophilic layers of the epidermis and dermis on skin absorption should be considered (influence of the partition step between the SC and viable epidermis/dermis). Moreover, dermatomed skin might be of particular interest for studying the fate of very lipophilic drugs (e.g., pesticides), due to their limited solubility in the rather hydrophilic viable epidermis and dermis, and for cosmetic products.

Design methodology of image processing systems has changed fundamentally in recent years. Previous methods based on heuristics have been replaced by deep learning neural networks trained for specific tasks. Using a trained network during inference requires high computing power. In low-power embedded systems, application-specific accelerators implemented on FPGAs are required. Different recognition tasks may require different network structures; therefore, adaptability of the network description is very important. The student should become acquainted with the frameworks for implementing deep learning neural networks on FPGAs and solve a specific recognition task.

Microfluidic devices make it possible to investigate interactions between individual yeast cells and to study their aging. This project offers an opportunity to develop microfluidic chambers and/or to answer biological questions (e.g., cell-cell interactions, aging, and the cell cycle at the single-cell level) with the help of such chambers.

The aim of this task is to investigate how adversarial attacks can be generated, detected, and recovered from in machine learning, in various image processing tasks such as classification, detection, and segmentation.

Our partners from Ukraine provided us with two different hydrogels, which are currently used for wound therapy for soldiers. These hydrogels are functionalized with silver nanoparticles for an antiseptic and antibiotic effect.

Our study aims to demonstrate the release and skin penetration of active components (silver nitrate) from the formulation in microfluidic diffusion chambers on human skin. We are looking for students to perform microfluidic studies.

Microbial communities are extremely important in our daily lives, both physiologically and from a biotechnological standpoint. In the Systems Biology group, we study the interactions of microbes, especially yeast strains. In our experimental work, the application of automated methods is crucial. We have liquid handling systems and a motorized microscope at our disposal. Our students can join our projects and take part in the development of high-throughput and automated protocols. We would be happy for students interested in robotics, automation, and high-throughput methods to join the group.

Microbial communities are extremely important in our daily lives, both physiologically and from a biotechnological standpoint. In the Systems Biology group, we study the interactions of microbes, especially yeast strains. In our experimental work, the application of automated methods is crucial. We have liquid handling systems and a motorized microscope at our disposal. Our students can join our projects and take part in the development of high-throughput and automated protocols. We would be happy for students interested in robotics, automation, and high-throughput methods to join the group.

Embedded systems readily recognize human faces and objects. However, interaction with humans or animals is a much more challenging, and rewarding, task. How can I, as a programmer, modify the behavior of a living organism? In this project, we develop interaction environments for animals and humans and train them on tasks that can report how well they see the world (visual acuity, depth perception, visual prediction). We have developed a virtual interaction environment and train subjects with minimal human intervention. In this project, you will learn how to interface hardware elements (Arduino, Raspberry Pi, depth cameras, EEG, reward systems), adapt the environment, design new interaction tasks, and analyze multimodal data.

Active dendritic mechanisms greatly enhance the information processing capabilities of single neurons and contribute to behaviorally relevant computations. In CA1 neurons, large dendritic Ca2+ spikes have been observed both in vitro and in vivo, but the precise biophysical mechanisms of their generation are unknown. In particular, it has been observed that Ca2+ spikes can be triggered by dendritic current injections in vitro, while somatic stimulation only leads to Na+ spikes. However, when somatic Na+ spikes are blocked, Ca2+ spikes can also be triggered from the soma, suggesting a complex and dynamical interaction between these two spike types. The goal of the project is to use simple and morphologically detailed biophysical models to explain these observations and provide a more realistic model of dendritic excitability in CA1 pyramidal cells. This work could reveal fundamental properties of dendrites with possible implications for biological and artificial neuronal networks.

Your primary responsibilities during the semester will include:

Data Collection and Integration: Assist clinicians in recording high-resolution data from patients and model systems, using techniques such as electrophysiology (ephys), functional ultrasound (fUS), and advanced clinical imaging (e.g., OCT).

Automated Data Processing: Develop a robust, automated pipeline to preprocess and clean data, prepare it for analysis, and extract key features using novel analysis methods.

Machine Learning Model Development: Implement, train, and optimize deep learning models designed for time-series classification, with the aim of distinguishing healthy signals from the earliest indicators of pathology.

Navigating in a novel environment or learning new information in a familiar one requires adaptive neuronal representations. Recent findings have identified biophysical mechanisms, behavioral time scale plasticity (BTSP) events, that can cause sudden changes in the spatial tuning of hippocampal neurons. However, it is unclear whether these BTSP events are necessary and sufficient for all forms of representational change. In particular, it is not known how repeated BTSP events during place field traversal influence the stability of existing place fields. Building on our recently developed scheme to detect potential BTSP events, the student will develop methods to analyze place field stability and predict lap-by-lap place cell activity.

Review the recent papers analyzing plasticity of hippocampal place fields, with special emphasis on the methods used to characterize lap-to-lap changes due to behavioral time-scale plasticity (BTSP) or representational drift. Develop novel analysis methods to predict place cell activity in a given lap based on activity history and behavioral variables. Validate and test the method on synthetic data. Finally, apply the analysis techniques to real data recorded in mice navigating in virtual reality environments.

Computational finance has been one of the most dynamically evolving research fields over the past 75 years. One of the most studied areas is the pricing and risk management of complex financial products, such as futures and options. The Nobel Prize-winning Black–Scholes option pricing model, introduced in the early 1970s, has received new momentum with the advancement of neural networks, machine learning, and other AI algorithms. Within the framework of the lab, students will have the opportunity to research these methods under professional supervision.

The high computational and memory requirements of vision transformer-based 3D SLAM and reconstruction models significantly limit their use on embedded and mobile platforms. The aim of this project is to reduce their memory footprint and computational requirements using quantization and other model optimization techniques (KV-cache compression, adaptive state updates), while minimizing the degradation of reconstruction and localization accuracy.

The project should investigate different numerical representations and quantization strategies, and evaluate their impact on memory usage, runtime, and accuracy. If necessary, additional optimization approaches may also be explored, such as using different numerical precision levels for individual model components or reducing the memory requirements of intermediate representations.

The longer-term goal is to make the selected models executable within resource constraints suitable for future deployment on go-kart, mobile, and drone platforms.

We know that chemogenetics allows us to switch specific neurons on or off, but what happens to the molecular machinery inside the tissue when we do? In this project, you will investigate the transcriptional aftermath of neuronal activation. Using bulk FLASH-seq (a high-sensitivity RNA sequencing method), you will analyze how targeted stimulation reshapes gene expression networks. You will look beyond simple activity markers to identify complex functional shifts, including:

Immediate early genes (IEGs): the first responders to neural activity.
Synaptic and structural proteins: how the brain physically remodels itself.
Neuroinflammatory signatures: investigating potential immune and glial responses to stimulation.

The aim of this project is to create a rotating platform for the evaluation of metasurfaces.

The aim of this project is to develop an RTAB-Map-based 3D SLAM and mapping pipeline for an NVIDIA Jetson NX embedded platform with 8 GB of memory installed on a drone. The current system uses Ubuntu 18.04; therefore, the proposed solution must be compatible with the existing software and hardware environment.

The goal of the project is to augment the capabilities of the widespread GPS/IMU-based self-motion estimation algorithms with information available from onboard camera images. The first part is to conduct a comprehensive literature review of the state of the art, and then to develop new algorithms for the task. It is also important to run tests to determine the statistical properties of these algorithms. The tests will be run either on videos rendered by the simulator or on flight videos. Preliminary algorithms, scientific papers, the simulator, and flight videos are available for the work.
Unmanned Aerial Vehicles (UAV), sensor fusion, collision avoidance, video and image processing

The goal of the project is to augment the capabilities of the widespread GPS/IMU-based self-motion estimation algorithms with information available from onboard camera images. The first part is to conduct a comprehensive literature review of the state of the art, and then to develop new algorithms for the task. It is also important to run tests to determine the statistical properties of these algorithms. The tests will be run either on videos rendered by the simulator or on flight videos. Preliminary algorithms, scientific papers, the simulator, and flight videos are available for the work.
Unmanned Aerial Vehicles (UAV), sensor fusion, collision avoidance, video and image processing

High-density laminar silicon probes have been widely used in rodent electrophysiology but were virtually absent from human research until recently. Clinical research devices have so far been able to detect single-unit activity (SUA) with poor spatial resolution (due to large inter-contact distances), or they have been limited to recording only the low-frequency component of extracellular neural activity (local field potentials). We pioneered a rare opportunity to test state-of-the-art silicon probes (namely the custom-designed, backside-thickened Neuropixels 1.0-S probes) in the operating room setting during resective surgeries for tumors or epilepsy, or during the implantation of a deep brain stimulator (DBS) in patients with Parkinson’s disease. These probes feature hundreds of closely packed active contact sites, enabling multichannel oversampling of extracellular SUAs. Spatiotemporally resolved morphoelectric properties of single units allow sophisticated clustering of neural cell types. Among several spatiotemporal features of cortical neurons, in this study we first focus on the so-called backpropagating action potential (bAP). Intracellular spikes can propagate backward along the somatodendritic axis. This phenomenon can also be detected extracellularly on spike-triggered averages (STAs) of the corresponding SUAs after spike sorting. These bAPs have been observed in rodents (using standard Neuropixels 1.0 probes) in vivo, with distinct spatiotemporal patterns across brain regions, putative cell types, and morphological orientations. By using algorithms and criteria from the published rodent literature, we report the first in vivo evidence for human neural action potential backpropagation from awake and anesthetized human patients. Our preliminary results on the detection of multichannel bAP patterns have shown consistency with the rodent findings, as we were able to identify canonical regular spiking (RS) putative principal cells expressing reliable bAPs, as well as fast-spiking (FS) putative interneurons that completely lack this spatial propagation. These results highlight the importance of high-density sampling in translational human in vivo recordings for disentangling cell types and morphological contributions to the local neocortical microcircuit.

Programming skills in at least one of the following languages: MATLAB, C++, Python (preferred, but it can also be learned on the fly). Advantage: interest in or previous experience with machine learning, statistics, modeling, or data science.

Keen interest in experimental neurobiology and electrophysiology, with a human clinical perspective.

Working language: English.

While 3D Gaussian Splatting (3DGS) has recently revolutionized novel view synthesis with its exceptional rendering speed and visual quality, its underlying scene representation relies on unstructured volumetric ellipsoids rather than explicit geometric surfaces. This project addresses the critical challenge of deriving accurate, continuous 3D surface meshes from trained 3DGS models. The primary objective is to investigate and implement state-of-the-art surface extraction techniques, bridging the gap between high-fidelity volumetric radiance fields and traditional polygon-based 3D geometry, which remains essential for physical simulations, collision detection, and standard rendering engines. To rigorously evaluate the effectiveness of the surface reconstruction methods, the research will utilize video data from two contrasting real-world scenarios. The first dataset will feature complex indoor environments captured via handheld mobile devices, testing the algorithms against variable lighting, occlusions, and detailed close-range geometries.

It is known from experiments that changes in body temperature affect electrical signals and metabolic processes in the brain. Microsystems implantable into neural tissue are manufactured for exploring these connections, and some of them are even suitable for freely behaving in vivo experiments. The applicant has to observe the functioning of the integrated thermal and electrophysiological sensors under experimental conditions, and has to calibrate them in vitro. Beyond designing and assembling the measuring system, the applicant has to complete documented functional tests of the implantable microtools. 

A variety of artificial intelligence (AI) methods are becoming standard tools in scientific research. An emerging trend involves the application of agentic AI systems, which promise to handle some or even all parts of the scientific process (including the identification of open problems, hypothesis generation, experimental design, data analysis, and publication) in a mostly autonomous manner, involving minimal human intervention. Simulation-based science also allows the actual experiments to be carried out by AI agents, which are capable of generating and executing the code that builds and simulates the model. The aim of this project is to test the applicability of this approach to computational neuroscience by using state-of-the-art agentic AI systems to design, execute, and evaluate simulation experiments involving detailed models of single neurons or simplified models of cortical networks.

This project focuses on developing advanced AI models to predict and understand gene expression within microbial communities, with implications for health diagnostics, ecosystem management, and synthetic biology. The project goal is to develop novel genomic language model-based solutions to understand the underlying gene-expression process in microbiomes. This is crucial, since obtaining transcriptome data is challenging both computationally and experimentally; therefore, delivering efficient models could greatly improve our capability of understanding engineered systems. The goal is to develop a new, multimodal AI model for gene prediction from a genomic context.

Understanding and reviewing the relevant literature; using an NGS pipeline; model training and evaluation