НИС Современный ML (ИИ24, 7 модуль): различия между версиями
Перейти к навигации
Перейти к поиску
Gazuev (обсуждение | вклад) м Gazuev переименовал страницу НИС «Современный ML» (ИИ24, 7 модуль) в НИС Современный ML (ИИ24, 7 модуль) |
Добавление информации о курсе |
||
| Строка 1: | Строка 1: | ||
==О курсе== | ==О курсе== | ||
Занятия проводятся в | Занятия проводятся в Zoom по вторникам в 19:40. | ||
==Контакты== | ==Контакты== | ||
Преподаватель: Петр Гринберг | |||
Преподаватель: | |||
==Материалы курса== | |||
Ссылка на плейлист курса на YouTube: [https://www.youtube.com/playlist?list=PLmA-1xX7IuzAM3T8NxmmnEjT72rim0HYJ YouTube Playlist] | |||
{| class="wikitable" | {| class="wikitable" | ||
! Номер занятия | |||
! Лекция / Разбираемая статья | |||
|- | |||
| 1 | |||
| Создание презентаций и рассказ докладов | |||
|- | |||
| rowspan="3" | 2 | |||
| ToolGen: Unified Tool Retrieval and Calling via Generation | |||
|- | |- | ||
! | | Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes | ||
|- | |- | ||
| | | Visual Instruction Tuning | ||
|- | |- | ||
| | | rowspan="3" | 3 | ||
| | | Scalable Diffusion Models with Transformers | ||
|- | |||
= | | Direct Preference Optimization: Your Language Model is Secretly a Reward Model | ||
|- | |||
| Graph Attention Networks | |||
|- | |||
| rowspan="3" | 4 | |||
| Voyager: An Open-Ended Embodied Agent with Large Language Models | |||
|- | |||
| VL2Lite: Task-Specific Knowledge Distillation from Large Vision-Language Models to Lightweight Networks | |||
|- | |||
| ReAct: Synergizing Reasoning and Acting in Language Models | |||
|- | |- | ||
| rowspan="3" | 5 | |||
| On Efficient Distillation from LLMs to SLMs | |||
|- | |- | ||
| | | Efficiently Modeling Long Sequences with Structured State Spaces | ||
|- | |- | ||
| | | Efficient Memory Management for Large Language Model Serving with PagedAttention | ||
|- | |- | ||
| | | rowspan="3" | 6 | ||
| Large Language Diffusion Models | |||
|- | |- | ||
| | | Toolformer: Language Models Can Teach Themselves to Use Tools | ||
|- | |- | ||
| | | Fast Inference from Transformers via Speculative Decoding | ||
|- | |- | ||
| | | 7 | ||
| Инструкция по написанию научных текстов | |||
|- | |- | ||
| | | rowspan="3" | 8 | ||
| ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs | |||
|- | |- | ||
| | | The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity | ||
|- | |- | ||
| | | Grad-CAM++: Improved Visual Explanations for Deep Convolutional Networks | ||
|- | |- | ||
| | | rowspan="2" | 9 | ||
| Предзащита ВКР | |||
|- | |- | ||
| | | Предзащита ВКР | ||
|- | |- | ||
| | | rowspan="3" | 10 | ||
| Предзащита ВКР | |||
|- | |- | ||
| | | Предзащита ВКР | ||
|- | |- | ||
| Предзащита ВКР | |||
|} | |} | ||
==Формула оценивания== | ==Формула оценивания== | ||
Накопленная = 0.4 * О_Квиз + 0.4 * О_Видео + 0.1 * О_Ревью + 0.1 * О_Мета-Ревью | |||
== | Оценка = 10 если Накопленная >= 5, иначе 0. | ||
О_Ревью и О_Мета-Ревью блокирующие. | |||
Текущая версия от 21:04, 19 апреля 2026
О курсе
Занятия проводятся в Zoom по вторникам в 19:40.
Контакты
Преподаватель: Петр Гринберг
Материалы курса
Ссылка на плейлист курса на YouTube: YouTube Playlist
| Номер занятия | Лекция / Разбираемая статья |
|---|---|
| 1 | Создание презентаций и рассказ докладов |
| 2 | ToolGen: Unified Tool Retrieval and Calling via Generation |
| Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes | |
| Visual Instruction Tuning | |
| 3 | Scalable Diffusion Models with Transformers |
| Direct Preference Optimization: Your Language Model is Secretly a Reward Model | |
| Graph Attention Networks | |
| 4 | Voyager: An Open-Ended Embodied Agent with Large Language Models |
| VL2Lite: Task-Specific Knowledge Distillation from Large Vision-Language Models to Lightweight Networks | |
| ReAct: Synergizing Reasoning and Acting in Language Models | |
| 5 | On Efficient Distillation from LLMs to SLMs |
| Efficiently Modeling Long Sequences with Structured State Spaces | |
| Efficient Memory Management for Large Language Model Serving with PagedAttention | |
| 6 | Large Language Diffusion Models |
| Toolformer: Language Models Can Teach Themselves to Use Tools | |
| Fast Inference from Transformers via Speculative Decoding | |
| 7 | Инструкция по написанию научных текстов |
| 8 | ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs |
| The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity | |
| Grad-CAM++: Improved Visual Explanations for Deep Convolutional Networks | |
| 9 | Предзащита ВКР |
| Предзащита ВКР | |
| 10 | Предзащита ВКР |
| Предзащита ВКР | |
| Предзащита ВКР |
Формула оценивания
Накопленная = 0.4 * О_Квиз + 0.4 * О_Видео + 0.1 * О_Ревью + 0.1 * О_Мета-Ревью
Оценка = 10 если Накопленная >= 5, иначе 0.
О_Ревью и О_Мета-Ревью блокирующие.