Translartion. Region: Russians Fedetion –
Source: Peter the Great St Petersburg Polytechnic University – Peter the Great St Petersburg Polytechnic University –
The spring cycle of seminars on artificial intelligence at the Polytechnic opened with a presentation of a project to improve the quality of plants and transition to green farming.
Alexander Fedotov, a leading researcher at the laboratory “Modeling of Technological Processes and Design of Power Equipment”, told the seminar participants about how artificial intelligence helps in processing multidimensional spatial data from remote sensing of natural and technical systems. Natural and technical systems are important for sustainable development, since they allow the efficient use of natural resources, minimizing damage to the environment.
Results were presented on the use of deep learning algorithms for recognizing objects in 3D scenes from laser scanning point clouds. This development is interesting because segmentation of 3D scenes is always a labor-intensive and non-trivial task.
The system for detecting phytosanitary threats based on artificial intelligence developed by a team of scientists allows determining the condition of plants and identifying their diseases at the earliest stage. To carry out the research, the scientists, together with colleagues from the All-Russian Research Institute for Plant Protection, created their own datasets of spectral portraits of diseased and healthy plants based on hyperspectral images.
Another relevant area is the analysis of transactions in blockchain networks. It plays an important role in the fight against money laundering. One of the key areas in this area is the classification of addresses, which allows identifying suspicious transactions and distinguishing between legitimate and illegal transactions. Using big data technologies, graph structure analysis, expert rules and machine learning methods (gradient boosting, such as LGBM, XGBoost, CatBoost, as well as interpretable AI methods (xAI SHAP), scientists were able to effectively track anomalous transactions. Through active learning, the model is constantly being improved. According to Alexander Fedotov, foreign solutions in this area are still inferior in efficiency, which emphasizes the need to develop domestic technologies for analyzing blockchain transactions.
Associates in the field of AI in pharmacology presented associate professors of the Higher School of Biomedical systems and technologies: the head of the nano-and microcapsulation of biologically active substances Alexander Timin and researchers of the laboratory Sergey Shipilovsky and Andrei Makashov. Scientists talked about world trends in solving the problem of manifestation of side effects from different drugs using the example of antitumor drugs. Currently, emphasis is on targeted use of drugs. Scientists of SPBPU, using a retrosynthetic analysis of large data arrays (Big Data), establish a dependence between the structure and biological activity. A trained neural network generates potential structures with the required properties and predicts the affinity of binding with targeted molecules. The proposed approach allows you to calculate the properties based on the structure, create training samples (more than 40,000 molecules), predict the structure of leading formations in the space of experimental samples. These decisions and the developed neural network filter, which monitors the effect of molecules on the body, significantly reduce temporary and material costs on preclinical studies. Answering the questions of the seminar about the reality of ambitions ten times to reduce the cost of new drugs to the market, young scientists replied that in the conditions of the possibilities that appeared with the departure of foreign companies from the Russian market and the interest of domestic manufacturers, their search technologies for the lead structure have already been studied by industrial partners and received approval. At the same time, Sergey Shipilovsky noted that their development is precisely the search for the most effective drugs, and not their creation, since artificial intelligence cannot be engaged in synthesis, it can only treat data, predict the properties of drugs.
Summing up the results of the seminar, the Head of the Department of Scientific Projects and Programs Natalia Leontyeva emphasized that cases involving industrial partners are of great interest, and invited to continue the topic at the next seminar, which will take place on March 26 at 14.00 in the Kapitsa Hall.
Please note: This information is raw content directly from the source of the information. It is exactly what the source states and does not reflect the position of MIL-OSI or its clients.