Posts

We use DLC (DeepLabCut) to label keypoint data and train the model using YOLO Pose. Since the data format required by YOLO Pose is different from DLC’s output, we need to convert it. With the help of AI today, it’s quite easy to generate code for format conversion. Here’s how we approached it: Interacted with AI to understand the differences between the DLC format and the YOLO format Defined the desired format for YOLO training data Asked AI to generate the conversion code based on the required format We found that: ...

last update : 2025-07-31 This is my side project, and I’m using it to document the logs during the prototyping phase. I’m trying to build a simple and low-cost camera-based system to help automate the detection of lameness in cows. I’m trying to answer these questions : Using current open-source methods, what level of accuracy can be achieved for specific problem detection? What are the typical optimization strategies, and what is the ROI for each? Does a deep understanding of fundamental principles, like CNNs, lead to improved accuracy? What specific aspects benefit from this knowledge? 1 Logs & Plans Week Phase Checkpoint Key Deliverables Acceptance Criteria 2 Data Annotation & Initial Model Training Completion of DLC labeling and initial YOLO Pose training 20–50 images annotated with DLC, initial YOLO Pose model trained Annotation error < 5 pixels;YOLO Pose mAP ≥ 70% 4 Lameness Detection Algorithm Development & Validation Implementation and initial video testing of lameness detection algorithm First version of lameness detection algorithm tested on videos Detection accuracy ≥ 70%;At least 10 video samples (5–8 seconds each) tested 6 Automated Data Processing Pipeline Setup Initial automation pipeline developed Automatic video capture → Upload to S3 → Automated processing & notification Pipeline success rate ≥ 90%;Processing delay ≤ 10 mins;Logging functionality enabled 8 System Integration & Final Validation Complete system test, optimization, and stability verification Comprehensive technical documentation and user manual completed Overall accuracy ≥ 75%;System stability ≥ 95%;Processing delay ≤ 10 mins;Clear and extensible documentation 1.1 Week 1~2 : Data Annotation & Initial Model Training Our goal is : ...

1 Basic concepts An LLM goes through two main phases(Lifecycle of an LLM): Training and Inference: Pre-training : it learns to predict the next word or token from large amounts of unlabeled text. Fine-tuning : The model is adapted to specific tasks or domains using labeled datasets. Post-training(Alignment) : it learns useful behaviors such as following instructions, tool use, and reasoning. Inference : The trained model responds to user inputs in real time without updating its internal parameters. Foundational Techniques that Underpin LLM Architecture : ...

I read a twitter from Andrej Karpathy: How to become expert at thing: iteratively take on concrete projects and accomplish them depth wise, learning “on demand” (ie don’t learn bottom up breadth wise) teach/summarize everything you learn in your own words only compare yourself to younger you, never to others So I write this for think about this topic. 1 Understand the basic rules Depth Feynman Technique : If you can’t explain it simply, you don’t understand it well enough. 2 from Beginner’s zone to Reality AK avoided a hard truth of the real world: Value is defined by relative position, not absolute effort. ...

I believe the same question applies to myself: how can I become a survivor in the world of the future? 1. Core Value: Create Value No matter the time or job, creating value is always important. The best jobs are those that help people and make a difference. 2. Key Principles: How to Grow That Value Technology drives progress : New tools and machines change the world. Use money to make more money : that’s the power of capital Use people to create more value : that’s the power of organization 3. Success Tips: Three Simple Rules Right time, right place, right people. Be brave, careful, and confident. Know people, work with people, lead people. 4. Learn from History 4.1 Lessons from Japan’s “Lost 20 Years” Japan’s “lost 20 years” saw the rise of affordable businesses like 7-Eleven, Uniqlo, MUJI, Daiso, and Saizeriya. Meanwhile, traditional giants like home appliances and cars declined. Japan missed the wave of computers, the internet, and mobile phones. 4.2 Career Choices of Top Chinese Graduates Top Chinese graduates didn’t all enter high-tech fields: ...

1 Khan Academy View 1.1 Year 4 Unit Topic ICAS Category 1 Place value Number & Arithmetic 2 Addition, subtraction, and estimation Number & Arithmetic 3 Multiply by 1-digit numbers Number & Arithmetic 4 Multiply by 2-digit numbers Number & Arithmetic 5 Division Number & Arithmetic 6 Factors, multiples and patterns Algebra & Patterns 7 Equivalent fractions and comparing fractions Number & Arithmetic 8 Add and subtract fractions Number & Arithmetic 9 Multiply fractions Number & Arithmetic 10 Understand decimals Number & Arithmetic 11 Plane figures Space & Geometry 12 Measuring angles Space & Geometry 13 Area and perimeter Measures & Units 14 Units of measurement Measures & Units 1.2 Year 5 Unit Topic ICAS Category 1 Decimal place value Number & Arithmetic 2 Add decimals Number & Arithmetic 3 Subtract decimals Number & Arithmetic 4 Add and subtract fractions Number & Arithmetic 5 Multi-digit multiplication and division Number & Arithmetic 6 Multiply fractions Number & Arithmetic 7 Divide fractions Number & Arithmetic 8 Multiply decimals Number & Arithmetic 9 Divide decimals Number & Arithmetic 10 Powers of ten Number & Arithmetic 11 Volume Measures & Units 12 Coordinate plane Space & Geometry 13 Algebraic thinking Algebra & Patterns 14 Converting units of measure Measures & Units 15 Line plots Chance & Data 16 Properties of shapes Space & Geometry 1.3 Year 6 Unit Topic ICAS Category 1 Ratios Number & Arithmetic 2 Arithmetic with rational numbers Number & Arithmetic 3 Rates and percentages Number & Arithmetic 4 Exponents and order of operations Number & Arithmetic 5 Negative numbers Number & Arithmetic 6 Variables & expressions Algebra & Patterns 7 Equations & inequalities Algebra & Patterns 8 Plane figures Space & Geometry 9 Coordinate plane Space & Geometry 10 3D figures Space & Geometry 11 Data and statistics Chance & Data 2 ICAS View 2.1 Number & Arithmetic No. Category Year Unit Topic K1 K2 1 Place Value & Number Concepts 4 1 Place value ✓ 2 Place Value & Number Concepts 5 1 Decimal place value 3 Place Value & Number Concepts 4 10 Understand decimals 4 Place Value & Number Concepts 5 10 Powers of ten 5 Place Value & Number Concepts 6 4 Exponents and order of operations 6 Arithmetic Operations 4 2 Addition, subtraction, and estimation 7 Arithmetic Operations 4 3 Multiply by 1-digit numbers 8 Arithmetic Operations 4 4 Multiply by 2-digit numbers 9 Arithmetic Operations 4 5 Division 10 Arithmetic Operations 5 5 Multi-digit multiplication and division 11 Arithmetic Operations 6 2 Arithmetic with rational numbers 12 Arithmetic Operations 6 5 Negative numbers 13 Fractions & Decimals 4 7 Equivalent fractions and comparing fractions 14 Fractions & Decimals 4 8 Add and subtract fractions 15 Fractions & Decimals 5 4 Add and subtract fractions 16 Fractions & Decimals 4 9 Multiply fractions 17 Fractions & Decimals 5 6 Multiply fractions 18 Fractions & Decimals 5 7 Divide fractions 19 Fractions & Decimals 5 2 Add decimals 20 Fractions & Decimals 5 3 Subtract decimals 21 Fractions & Decimals 5 8 Multiply decimals 22 Fractions & Decimals 5 9 Divide decimals 23 Ratios & Percents 6 1 Ratios 24 Ratios & Percents 6 3 Rates and percentages 25 Patterns, Factors & Multiples 4 6 Factors, multiples and patterns 2.2 Algebra & Patterns No. Category Year Unit Topic K1 K2 1 Algebra & Patterns 4 6 Factors, multiples and patterns 2 Algebra & Patterns 5 13 Algebraic thinking 3 Algebra & Patterns 6 6 Variables & expressions 4 Algebra & Patterns 6 7 Equations & inequalities 2.3 Measures & Units No. Category Year Unit Topic K1 K2 1 Measures & Units 4 13 Area and perimeter 2 Measures & Units 4 14 Units of measurement 3 Measures & Units 5 11 Volume 4 Measures & Units 5 14 Converting units of measure 2.4 Space & Geometry No. Category Year Unit Topic K1 K2 1 Space & Geometry 4 11 Plane figures 2 Space & Geometry 4 12 Measuring angles 3 Space & Geometry 5 12 Coordinate plane 4 Space & Geometry 5 16 Properties of shapes 5 Space & Geometry 6 8 Plane figures 6 Space & Geometry 6 9 Coordinate plane 7 Space & Geometry 6 10 3D figures 2.5 Chance & Data No. Category Year Unit Topic K1 K2 1 Chance & Data 5 15 Line plots 2 Chance & Data 6 11 Data and statistics

Understanding Cognitive Levels in ICAS : Cognitive Level Description Example in ICAS Context Recall / Remembering Recall facts, terms, units, definitions without manipulation Know that 1 kilometre = 1000 metres Understanding Explain or interpret facts; classify, describe, or summarize Explain the meaning of a pie chart Applying Use information in new or practical situations Calculate the area of a rectangle given its dimensions (Pattern Recognition) Identify regularities, trends, or rules among data or sequences Predict the next number in a sequence: 2, 4, 8, ? Reasoning / Analyzing Break down problems, infer relationships, or make logical deductions Determine the rule of a pattern: 3, 5, 8, 12, 17, ? Creating / Modeling Combine knowledge to generate new ideas or models Create an equation to model a real-world relationship Evaluating Make judgments based on criteria, often comparing alternatives Compare which measurement estimate is most reasonable ICAS Mathematics – Cognitive Abilities Focus by Module : ...

1. Value Creation on Existing Software Unified chat interface reduces context-switching Automated, trigger-based workflows (“say once, do always”) Knowledge-driven outputs with continuous learning loops Scalable plugin ecosystem for third-party integrations The real tool is simply an interface for converting natural language. 2. Key Application Technologies Natural Language Understanding & Dialogue Management Workflow orchestration (triggers, branching, retries) Retrieval-Augmented Generation (RAG) for content accuracy Plugin invocation framework (OpenAPI/gRPC, sandboxing) Low-code visual builders (flow & dialogue editors) 3. Key Foundational Technologies Transformer-based LLMs and fine-tuning pipelines Vector databases and similarity search engines Distributed workflow engines (Temporal, Airflow) Containerized microservices (Kubernetes, Docker) API gateways and service meshes (Istio, Envoy) Observability stacks (Prometheus, Jaeger) 4. Current Challenges NLU robustness, multi-turn context, domain adaptation Ensuring vector index freshness and retrieval consistency Maintaining third-party connectors amid API churn End-to-end observability and fault-recovery at scale Data privacy, compliance, and secure model/data handling Cost-efficient inference and infrastructure resource control

This blog is still a work in progress; I’ll continue to update it. last update : 2025-05-20 1. Foundations of Machine Learning Core concepts: supervised vs. unsupervised learning, key algorithms Neural networks basics: perceptron → deep feed-forward → CNNs / RNNs Introduction to NLP & Transformer: attention mechanism, encoder–decoder architectures To grasp the basic concepts in about two hours, you can choose Understanding LLMs from Scratch Using Middle School Math. If you have more time, you might explore 3Blue1Brown’s series on neural networks. ...

“All education is about learning to think. It’s about learning to problem solve. And whether that was 30 years ago, now or 30 years from now. I think the technology changes over time, but if you can learn how to solve problems, you are going to do great things.” — Lisa Su (AMD CEO) The world is changing faster than ever before, thanks to the rapid development of AI and other new technologies. In this environment, the most important thing we can learn is not any specific tool or language, but the ability to solve problems. ...