The Skoltech Summer School of Machine Learning (SMILES-2025) is a 12-day intensive course focused on generative methods in AI. It is organized by Skoltech in collaboration with the Harbin Institute of Technology (HIT), China. The program will take place from July 14 to July 27, 2025.

SMILES-2025 will welcome 100 participants from all over the world attending in person, while up to 300 participants will join online. All attendees will have access to cutting-edge lectures, hands-on workshops, and hackathons designed to tackle real-world industry challenges. The school aims to bring together talented young researchers from Russia and China. One of the key goals of SMILES-2025 is to foster international collaboration and knowledge exchange. Participants will gain practical experience in advanced AI techniques, including generative models, multimodal methods, large language models, multi-agent systems. The school also offers professional networking, career development, mentoring from industry experts, and a cultural excursion for a well-rounded experience.

➔ Large language models

➔ Multimodal approaches

➔ Multi-agent systems

➔ Generative approaches

➔ Self-learning methods

➔ Safe AI

The summer school will be held at the Faculty of Computing of the Harbin Institute of Technology (HIT) in Harbin, China.

SMILES-2025 is designed for students, graduate students, early-career researchers, and AI enthusiasts from Russia and China who are actively engaged in advancing modern machine learning through innovative research and applications.

Potential participants are required to submit the following set of documents and complete a small test:

➔ CV
Describing your: education level, participation in scientific work, scientific events and teaching activities. Achievements: published articles, relevant work experience, topics of interest, thesis topics with the place of graduation. ML/DL tool skills and proficiency. 1-2 pages.

➔ Motivation letter;
Up to 1000 words long in which it is necessary to disclose: motivation to participate in school, ML/DL experience, if available, school expectations and future career plans

➔ Test with 10 questions on relevant calculus, statistics, machine learning and deep learning;

➔ Review Presentation: slides (pdf) + video recording (mp4/avi/mov/webm). Video recording is optional but it will give you extra points.

There are three possible options for a review:

• A description of your own ML/DL research
• A reproduction of the results of an article from one of the school's topics over the past 5 years. Adding ideas for improvement will be considered an additional plus to your application.
• A review of at least 3 articles on one of the school's topics

The video should last 5-10 minutes; ensure it captures your face, voice, and the slides. Upload the video to a cloud drive and don't forget to give us access to view it.

Review requirements
Articles for review and reproduction should be from A* rank conferences (NIPS, ICML, CVPR, ICCV, ECCV, AAAI, ACL, EMNLP, etc).
The presentation should include:

1. Introduction
2. Problem statement
3. Methods
4. The analytical table
5. Link to the repository if the article is being reproduced

Requirements for the analytical table:

• Analysis of trends on the chosen topic over the past 5 years based on articles, patents and open sources (NeurIPS, CVPR, etc.).
• Comparison of datasets (size, domain, preprocessing) and justification of their choice.
• Evaluation metrics with an explanation of why some metrics are used more often than others and what problems they may have.
• Analysis of SOTA models: analysis of efficiency on different tasks, complexity of implementation.
• Comparison of implementations in repositories (Hugging Face, GitHub, Papers with Code): pros and cons, configuration difficulties.
• You can use the first table here as an example.

Authenticity is crucial

We strongly discourage using large language models (LLMs) to compose your application materials. Instead, write from your heart, in your own voice, to capture your one-of-a-kind perspective and passion. We read hundreds of applications and look for submissions that demonstrate genuine enthusiasm and originality. While LLMs can be tempting, they tend to produce generic content with little value — ultimately reducing your chances of success. Please note that we will employ an LLM detector to verify the integrity of all submitted materials.

Participants who have passed the selection process must prepare and present a poster with the results described in the abstract at the conference. This is a long-standing tradition of summer schools: we want to introduce all school participants to your work. Only with a poster, you will get access to SMILES.
➔ (PDF) example of a poster made in LATEX
➔ (PDF) example of a poster made in POWERPOINT

Each day will consist of three lectures or seminars, followed by time for working on projects and for relaxation. The schedule will be published later.