Project 1: Guideline TensorFlow for AI – Get to Know TensorFlow

Epoch	Learning Objective	Core Topics / Activities	Expected Outcome
1. Introduction to TensorFlow & Setup	Understand TensorFlow basics and installation	TensorFlow installation, GPU setup, tensor operations, eager execution	TensorFlow environment ready and basic operations understood
2. Tensor, Variables & Operations	Learn about TensorFlow data structures	Tensors, Variables, Rank, Shape, Broadcasting, Basic math ops	Ability to manipulate tensors effectively
3. Automatic Differentiation & GradientTape	Understand backpropagation in TensorFlow	GradientTape, computing gradients, optimization basics	Ability to compute and apply gradients manually
4. Linear Regression & Optimizers	Build and train a simple regression model	tf.keras basics, loss functions, GradientDescentOptimizer	Implemented first end-to-end ML model

Evaluation – Project 1:

Evaluation criteria	Requirement	Weight (%)
A. Understand Tensor & Gradient	Explain the role of Tensor and Gradient in training	20%
B. Reasonable model structure	The model has full Input, Hidden, Output, Activation	25%
C. Reasonable training results	Loss gradually decreases, Accuracy/Loss report correctly	25%
D. Explain results & clean code	Clear code, with comments, understand the meaning of the functions	20%
E. Report results	Has a reasonable Loss graph or output description	10%
Meets the requirements when: Score ≥ 70% and the model runs successfully

Project 2: Guideline Neural Network from Scratch in TensorFlow

Understand Neural Network, CNN, and Transfer Learning architecture.
Process image data and train Dogs vs Cats classification model.
Evaluate, optimize, and visualize results.

Epoch	Learning Objective	Core Topics / Activities	Expected Outcome
5. Building Neural Networks with Keras	Create neural network models using Keras API	Sequential model, Dense layers, activation functions, compile & fit	Built a working neural network for basic classification
6. Image Data Preparation	Handle real-world image datasets	ImageDataGenerator, rescaling, augmentation, splitting data	Dataset ready for training CNN
7. Convolutional Neural Network (CNN) Basics	Build and train CNNs for image recognition	Conv2D, MaxPooling2D, Flatten, Dropout	Built CNN achieving 80–85% accuracy on training data
8. Transfer Learning with Pretrained Models	Improve accuracy and efficiency	VGG16, MobileNetV2, feature extraction, fine-tuning	Model accuracy improved >90%, reduced training time

Evaluation – Project 2:

Evaluation Criteria	Requirement Description	Weight (%)
A. Data Pipeline	Data is processed correctly (rescale, augment, split appropriately)	20%
B. Model Architecture	CNN model has reasonable structure, standard activation	25%
C. Training & Validation	Successful training, clear accuracy/loss graph	25%
D. Transfer Learning Implementation	Correctly apply pre-trained model techniques	20%
E. Reporting & Analysis of results	Describe why the model works well/badly	10%
Meets requirements when: Accuracy ≥ 85% and score ≥ 70% total

Project 3: Guideline Deploy Models with TensorFlow Serving and Flask

Learn how to save, load, and deploy a TensorFlow model.
Create a REST API using Flask to receive images and return results.
Get familiar with TensorFlow Serving (Docker).

Epoch	Learning Objective	Core Topics / Activities	Expected Outcome
10. Model Saving & Loading	Save and reuse models	- `model.save()` and `tf.keras.models.load_model()` - SavedModel, H5, TF Lite formats	Save/restore models successfully
11. Flask API for Prediction	Create API to serve models	- Flask server - Endpoint `/predict` to upload images - Predict and return JSON results	Flask API runs and returns accurate results
12. TensorFlow Serving & Optimization	Optimize & deploy real products	- TensorFlow Serving via Docker - Quantization, Pruning - Real-time inference demo	Models can run outside the dev environment

Evaluation – Project 3:

Deployment Functionality (40%) – Flask API correctly predicts and handles requests

Integration & Design (30%) – Clean integration between TensorFlow, Flask, and frontend

Testing & Reporting (30%) – Demonstrated end-to-end tests and user documentation

I. Systems of Equations and Sentences

1. “Linear algebra is systems of linear equations.”

Trong ML, hầu hết các mô hình đều được mô tả bằng linear algebra:

Input (features) ⇒ vector
Parameters (weights) ⇒ vector/matrix
Predictions ⇒ kết quả từ linear equations

Ví dụ:

\[y = w_1 x_1 + w_2 x_2 + \cdots + w_n x_n + b\]

Đây chính là một systems of linear equations (mô hình linear regression cơ bản).

2. “When you think of equations, you think of sentences.”

Một equation giống như một sentence trong ngôn ngữ: nó truyền đạt một thông tin về dữ liệu.
Trong ML: mỗi phương trình biểu diễn một constraint (ràng buộc) hoặc một data point.

Ví dụ:

Phương trình (Equation): 2x + y = 5
Sentence: “Two apples and one banana cost 5 dollars.”
Trong ML: Đây là một dữ liệu huấn luyện (x,y).

3. “Sentences that are giving you information about things in the world.”

Trong ngôn ngữ: mỗi câu mô tả một sự thật.
Trong ML: mỗi data point (sample) cũng là một “câu” về thế giới thực.

Ví dụ::

Sentence 1: “Nam is taller than Lan.”
Sentence 2: “Lan is 160 cm tall.”

⇒ We can infer: “Nam is taller than 160 cm.”

\[\begin{cases} A > B \\ B = 160 \end{cases} \quad \Rightarrow \quad A > 160\]

4. “Systems of sentences behave a lot like systems of equations.”

Ghép nhiều câu lại, ta hiểu rõ hơn bức tranh toàn cảnh.
Trong ML: nhiều data points (training set) tạo nên một system of equations mà mô hình phải giải để tìm tham số phù hợp.

Ví dụ (linear regression với 2 điểm dữ liệu):

\[\begin{cases} 2x + y = 5 \\ x - y = 1 \end{cases}\]

Giống như trong ML, bạn có nhiều mẫu huấn luyện, và bạn cần tìm nghiệm (weights) phù hợp với tất cả.

5. “Systems of sentences combine themselves to give you more information.”

Khi các câu kết hợp, ta có thể suy ra điều mới.
Trong ML, khi nhiều data points kết hợp, mô hình tổng quát hóa để đưa ra dự đoán cho dữ liệu chưa thấy.

Ví dụ:

Training data (sentences):
- “If it rains, the ground gets wet.”
- “It is raining today.”

⇒ Kết hợp: “The ground will be wet today.”

Trong ML: mô hình học từ dữ liệu, rồi dự đoán cho input mới.

6. Example for system of sentences

Hệ thống này chứa nhiều thông tin như sentences và được gọi là a complete system.

The sentences lặp lại và do đó hệ thống này được gọi là dư thừa (redundant).

Hệ thống này được gọi là contradictory system vì con chó không thể vừa đen vừa trắng cùng một lúc, hãy nhớ rằng chúng ta có một con chó và nó chỉ có thể có một màu.

Non-singular: hệ có nghiệm duy nhất (complete system). Singular: hệ có vô số nghiệm (redundant system) hoặc hệ vô nghiệm (contradictory).

II. System of Linear Equations

1. From Sentences to Equations

Quy tắc chuyển đổi

S1: Xác định thực thể (danh từ) → biến số cần tìm.
S2: Chọn kiểu biến:
- Số thực / nguyên → biến x,y.
- Thuộc tính phân loại (màu, trạng thái) → mã hóa nhị phân (one-hot, 0–1).
- Logic/Boolean → biến nhị phân (0 hoặc 1).
S3: Dịch quan hệ (động từ, số lượng) thành ràng buộc toán học:
- bằng (=),
- nhiều hơn / ít hơn (≥,≤)
- “một trong hai” → tổng bằng 1.
S4: Thêm ràng buộc miền giá trị (ví dụ: số nguyên, không âm, nhị phân).
S5: Phân tích hệ:
- Đủ phương trình, det(𝐴) ≠ 0 → nghiệm duy nhất.
- Thiếu phương trình hoặc ràng buộc dư thừa → vô số nghiệm.
- Mâu thuẫn → vô nghiệm.

1.1. Ví dụ 1

Câu: “Between the dog and the cat, one is black”

\[\begin{cases} d_B = 1, \text{ If the dog is black, else 0} \\ d_O = 1, \text{ If the dog is organce} \\ c_B, c_O \text{ Similar for cat} \end{cases} \quad \Rightarrow \quad \begin{cases} d_B + d_O = 1 \\ c_B + c_O = 1 \end{cases}\]

Câu: Each animal has only one color. $\text{Each animal has only one color.} \quad \Rightarrow \quad d_B + c_B = 1$

\[\quad \Rightarrow \quad \begin{cases} d_B + d_O = 1 \\ c_B + c_O = 1 \\ d_B + c_B = 1 \end{cases}\]

1.2. Ví dụ 2

Câu: “The price of an apple an panana is $10”

Biến số
- a = giá táo.
- b = giá chuối.
Phương trình $a + b = 10$

Một phương trình, hai ẩn → vô số nghiệm.

Ví dụ nghiệm:

a = 6, b = 4
a = 3.5, b = 6.5
a = 10, b = 0
… any pair summing to 10.

Bổ sung câu để đủ nghiệm duy nhất: “The apple costs twice the banana.” → a = 2b

\[\begin{cases} a + b = 10\\[4pt] a - 2b = 0 \end{cases} \quad\Rightarrow\quad \text{solve: } 2b + b = 10 \Rightarrow b=\tfrac{10}{3},\; a=\tfrac{20}{3}.\]

1.3. Quizzes

Quizz 1

You go two days in a row and collect this information:

Day 1: You bought an apple and a banana and they cost $10.
Day 2: You bought an apple and two bananas and they cost $12.

Question: How much does each fruit cost?

Solution:

Day 1: You bought an apple and a banana and they cost $10.

Day 2: You bought an apple and two bananas and they cost $12.

Solution: An apple costs $8, a banana costs $2.

Quiz 2

You go two days in a row and collect this information:

Day 1: You bought an apple and a banana and they cost $10.
Day 2: You bought two apples and two bananas and they cost $20.

Question: How much does each fruit cost?
Day 1: You bought an apple and a banana and they cost $10.

Day 2: You bought two apples and two bananas and they cost $20.

Quiz 3

You go two days in a row and collect this information:

Day 1: You bought an apple and a banana and they cost $10.
Day 2: You bought two apples and two bananas and they cost $24.

Question: How much does each fruit cost?

Project 1: Guideline TensorFlow for AI – Get to Know TensorFlow

Evaluation – Project 1:

Project 2: Guideline Neural Network from Scratch in TensorFlow

Evaluation – Project 2:

Project 3: Guideline Deploy Models with TensorFlow Serving and Flask

Evaluation – Project 3:

I. Systems of Equations and Sentences

1. “Linear algebra is systems of linear equations.”

2. “When you think of equations, you think of sentences.”

3. “Sentences that are giving you information about things in the world.”

4. “Systems of sentences behave a lot like systems of equations.”

5. “Systems of sentences combine themselves to give you more information.”

6. Example for system of sentences

II. System of Linear Equations

1. From Sentences to Equations

1.1. Ví dụ 1

1.2. Ví dụ 2

1.3. Quizzes

Quizz 1

Quiz 2

Quiz 3

2.2. Systems of equations

2.3. A linear equation

2.4. From Linear equation to line

2.5 Systems of equations as lines

A geometric notion of singularity

2.6 Singular vs nonsingular matrices

Linear dependence and independence

The determinant - Linear dependence between rows

The determinant

Determinant and singularity