AI and Machine Learning
前往频道在 Telegram
Learn Data Science, Data Analysis, Machine Learning, Artificial Intelligence, and Python with Tensorflow, Pandas & more! Buy ads: https://telega.io/c/machine_learning_courses
显示更多📈 Telegram 频道 AI and Machine Learning 的分析概览
频道 AI and Machine Learning (@machine_learning_courses) 英语 语言赛道中的 是活跃参与者。目前社区聚集了 96 443 名订阅者,在 教育 类别中位列第 1 518,并在 印度 地区排名第 2 761 位。
📊 受众指标与增长动态
自 невідомо 创建以来,项目保持高速增长,吸引了 96 443 名订阅者。
根据 15 十一月, 2026 的最新数据,频道保持稳定运转。过去 30 天订阅人数变化为 210,过去 24 小时变化为 7,整体触达仍然可观。
- 认证状态: 未认证
- 互动率 (ER): 平均受众互动率为 8.22%。内容发布后 24 小时内通常能获得 2.29% 的反应,占订阅者总量。
- 帖子覆盖: 每篇帖子平均可获得 7 932 次浏览,首日通常累积 2 209 次浏览。
- 互动与反馈: 受众积极参与,单帖平均反应数为 11。
- 主题关注点: 内容集中在 learning, llm, linkedin, linux, udemy 等核心主题上。
📝 描述与内容策略
作者将该频道定位为表达主观观点的平台:
“Learn Data Science, Data Analysis, Machine Learning, Artificial Intelligence, and Python with Tensorflow, Pandas & more!
Buy ads: https://telega.io/c/machine_learning_courses”
凭借高频更新(最新数据采集于 16 十一月, 2026),频道始终保持新鲜度与高覆盖。分析显示受众积极互动,使其成为 教育 类别中的关键影响点。
96 443
订阅者
+724 小时
无数据7 天
+21030 天
帖子存档
96 443
🔅 Azure AI Services Essential Training
📝 Get an overview of the basic cognitive services available in Azure: vision, speech, language, language, search, content safety, translator, and document intelligence.
🌐 Author: Sahil Malik
🔰 Level: Intermediate
⏰ Duration: 2h 19m
📋 Topics: Microsoft Azure
🔗 Join Artificial intelligence for more courses
96 443
AI doesn't slow down. Neither does the news.
New models from OpenAI and Anthropic. Startups appearing out of nowhere. Security incidents, major breakthroughs, and billion-dollar deals.
There's always something happening. And keeping up with all of it is getting harder.
TIKRD AI NEWS brings you the AI stories worth knowing about.
Stay in the loop.
Follow TIKRD AI NEWS
96 443
🔥 Google has taught AI to refine its own rules—even when humans themselves aren’t sure what they want.
Google Research has introduced "Agile Deliberation," a system for building AI classifiers that must navigate subjective concepts.
For instance, how do you teach a neural network to distinguish between healthy and unhealthy food? A salad clearly fits, but a burger doesn’t—yet what about a fruit smoothie packed with sugar?
Traditional approaches require writing detailed rules in advance. Agile Deliberation, by contrast, identifies ambiguous examples and asks a human to decide how they should be categorized.
Based on these human responses, an LLM automatically refines the instructions for the vision model—all without the need for retraining or adjusting model weights.
The study utilized Gemini 2.5 Pro and Flash.
Experimental results:
- Up to an 11% improvement in F1 score compared to a baseline classifier.
- A 7.5% higher F1 score compared to automated category partitioning.
- A more than 3% higher F1 score compared to manual configuration.
- All 18 study participants preferred this new approach over manual prompt editing.
Google has released the source code and provided a demonstration in Colab.
Research: https://research.google/blog/agile-deliberation-how-ai-helps-humans-define-the-gray-areas/
96 443
FREE MIT books on AI and Machine Learning: 📚🤖
1. Foundations of Machine Learning cs.nyu.edu/~mohri/mlbook/
2. Understanding Deep Learning udlbook.github.io/udlbook/
3. Introduction to Machine Learning Systems ❯ Vol 1: mlsysbook.ai/vol1/assets/do ❯ Vol 2: mlsysbook.ai/vol2/assets/do
4. Algorithms for ML algorithmsbook.com
5. Deep Learning deeplearningbook.org
6. Reinforcement Learning andrew.cmu.edu/course/10-703/
7. Distributional Reinforcement Learning direct.mit.edu/books/oa-monog
8. Multi Agent Reinforcement Learning marl-book.com
9. Agents in the Long Game of AI direct.mit.edu/books/oa-monog
10. Fairness and Machine Learning fairmlbook.org
11. Probabilistic Machine Learning
❯ Part 1 : probml.github.io/pml-book/book1
❯ Part 2 : probml.github.io/pml-book/book2
96 443
⭐️ The Odyssey (2026) [BluRay Version]
Odysseus, the legendary King of Ithaca, embarks on a long and perilous journey home following the Trojan War. Throughout his voyage, he is forced to confront the whims of gods, mythological monsters, and trials that stretch both his cunning and his humanity to the breaking point.
🔒 VIP Exclusive
This documentary is only visible to vip members.
🔗 Subscribe first: VIP Channel
🔗 Then view content: Link
For more information go to: The Premium Vault
96 443
🔅 PREMIUM CHANNELS
-◦-◦--◦--◦-◦--◦--◦-◦--◦--◦-◦--◦-
🔰 Web Development
-◦-◦--◦--◦-◦--◦--◦-◦--
216k| 🔰 Linkedin Learning
143k| 🔰 Zero To Mastery
133k| 🔰 Web Development
-◦-◦--◦-
125k| 🔰 Learn Python 3
110k| 🔰 Boosmodz Box
096k| 🔰 Machine Learning
-◦-◦--◦-
096k| 🔰 Learn JavaScript
071k| 🔰 Artificial Intelligence
070k| 🔰 Data Analysis and Databases
-◦-◦--◦-
068k| 🔰 Linux and DevOps
053k| 🔰 Business and Finance
051k| 🔰 100 Days of Python
-◦-◦--◦-
050k| 🔰 AI Tools
043k| 🔰 Udemy Learning
041k| 🔰 Best Telegram Channels
-◦-◦--◦-
041k| 🔰 ZTM Courses
039k| 🔰 Mobile Apps
035k| 🔰 Soft Skills
-◦-◦--◦-
035k| 🔰 Linkedin Learning Courses
034k| 🔰 Codedamn Courses
030k| 🔰 Crypto Tutorials
-◦-◦--◦-
030k| 🔰 Coding Interview
027k| 🔰 Agentic AI Coding
025k| 🔰 The Coding Space
-◦-◦--◦--◦-◦--◦--◦-◦--
🔰 Add Your Channel
-◦-◦--◦--◦-◦--◦--◦-◦--◦--◦-◦--◦-
🔰 2hrs on top & 8hrs in channel!
96 443
🔅 PREMIUM CHANNELS
-◦-◦--◦--◦-◦--◦--◦-◦--◦--◦-◦--◦-
🔰 Web Development
-◦-◦--◦--◦-◦--◦--◦-◦--
216k| 🔰 Linkedin Learning
143k| 🔰 Zero To Mastery
133k| 🔰 Web Development
-◦-◦--◦-
125k| 🔰 Learn Python 3
096k| 🔰 Machine Learning
096k| 🔰 Learn JavaScript
-◦-◦--◦-
071k| 🔰 Artificial Intelligence
070k| 🔰 Data Analysis and Databases
068k| 🔰 Linux and DevOps
-◦-◦--◦-
053k| 🔰 Business and Finance
051k| 🔰 100 Days of Python
050k| 🔰 AI Tools
-◦-◦--◦-
042k| 🔰 Udemy Learning
041k| 🔰 Best Telegram Channels
041k| 🔰 ZTM Courses
-◦-◦--◦-
039k| 🔰 Mobile Apps
035k| 🔰 Linkedin Learning Courses
035k| 🔰 Soft Skills
-◦-◦--◦-
034k| 🔰 Codedamn Courses
030k| 🔰 Crypto Tutorials
030k| 🔰 Coding Interview
-◦-◦--◦-
027k| 🔰 Agentic AI Coding
025k| 🔰 The Coding Space
-◦-◦--◦--◦-◦--◦--◦-◦--
🔰 Add Your Channel
-◦-◦--◦--◦-◦--◦--◦-◦--◦--◦-◦--◦-
🔰 2hrs on top & 8hrs in channel!
96 443
🔎 Xiaomi has released an open-source model for document parsing with a size of just 0.8 billion parameters.
Xiaomi-OCR-0 integrates text recognition, document structure parsing, key information extraction, and question-answering based on document content.
In other words, the model’s scope goes beyond simple "transcribing text from an image"—it helps transform documents into data that can be used for further processing.
According to the developers:
- It scores 95.24 points on Real5-OmniDocBench.
- It achieves an average of 83.2 points across five benchmarks focused on text-recognition-based question-answering.
- It ranks first among models with fewer than 1 billion parameters for document parsing quality on OmniDocBench v1.6.
The model weights, inference code, an interactive demo, an MCP (Model Context Protocol) service, and an "Agent Skill" have all been made publicly available. The latter two components allow document processing capabilities to be integrated into AI agent workflows.
Its compact size makes the model an attractive option for deployment within your own infrastructure. However, performance on specific types of scans, languages, or complex tables should be verified independently—benchmark results do not guarantee accuracy for every possible document.
🤗 Hugging Face: https://huggingface.co/xiaomi-research/Xiaomi-OCR-0
🔗 GitHub: https://github.com/xiaomi-research/Xiaomi-OCR-0
📄 Demo: https://huggingface.co/spaces/loongwayX/Xiaomi-OCR-0-demo
96 443
Google has introduced a "sleep" phase for language models—a distinct stage during which the AI consolidates new knowledge and trains itself using its own generated examples.
A key challenge in continual learning is "catastrophic forgetting": as a model acquires new information, it may lose previously learned skills. Furthermore, information held within a model's context window does not automatically become part of its long-term, permanent memory.
The SLEEP framework alternates between two modes:
Wakefulness: The model interacts with incoming data and stores information in its short-term memory.
Sleep: The intake of new data is paused. The model activates specific "Mixture-of-Experts" (MoE) memory blocks and integrates the new knowledge into their weights, minimizing interference with previously consolidated information. Once this process is complete, the short-term memory is cleared.
There is also a "dreaming" phase, where random combinations of "experts" generate training examples. The system then selects the most useful of these examples to further refine the model's learning.
Experimental results from the team include:
- BABILong: The model can process information from sequences of up to 10 million tokens by consolidating it directly into its parameters.
- Sequential translation training: The model shows virtually no performance degradation in languages learned previously.
- Mathematics: The model achieved 79.2% accuracy on AIME 2024 and 69% on AIME 2025.
In this context, "sleep" functions as a controlled fine-tuning cycle, governed by rules established by the researchers. The model's underlying architecture and code remain fixed.
https://research.google/blog/the-sleep-framework-a-new-way-for-llms-to-retain-knowledge-in-continual-learning/
