This is an artificial intelligence teaching website aimed at zero-foundation learners. The author positions it as an accessible, down-to-earth AI tutorial that advocates happy learning and adds humor to the teaching to help learners persist to the end. The website introduction says that traditional textbooks have a relatively high threshold, usually requiring knowledge of programming, advanced mathematics, statistics, and so on, while this website strives to require no advanced mathematics; as long as one has attended high school, one can learn.
The website content covers many aspects of artificial intelligence, including basic principles, large models, agents, AIGC, autonomous driving, stock prediction, financial analysis, and some niche special applications. The author states that the website integrates knowledge accumulated over more than twenty years and provides practical programming code after each section.
From the course catalog, the website is mainly divided into the following sections:
-
Fundamentals of Artificial Intelligence: including what a neural network is, how to input data into a neural network, how a neural network makes predictions, how the network learns, computational graphs, partial derivatives of logistic regression, vectorization, basic Python syntax, Python flow control, Python container data structures, Python functions, a quick introduction to Python libraries, etc., and is accompanied by practical programming content in 'Teaching You to Write Your First Artificial Intelligence Program.'
-
Neural Networks: covering shallow neural networks, forward propagation, backpropagation, activation functions, partial derivatives of activation functions, random parameter initialization, nonlinearity and activation functions, deep neural networks, matrix dimension checking, parameters and hyperparameters, supervised learning neural networks, and practical programming for building deep neural networks.
-
Additional Knowledge: including scalars, vectors, matrices, and tensors, an in-depth understanding of matrices, norms, calculus, classical calculus, limit calculus, partial derivatives, directional derivatives, probability theory, conditional probability, information theory, conditional entropy, mutual information, relative entropy (KL divergence), cross-entropy, and test questions.
-
Practical Optimization: including dataset configuration, underfitting and overfitting, regularization, dropout, data augmentation, input feature normalization, vanishing and exploding gradients, judging network bugs, H5 files, as well as practical programming such as parameter initialization, regularization, and gradient checking; the optimization algorithms section includes mini-batch, exponentially weighted averages, momentum gradient descent, RMSprop, the Adam optimization algorithm, learning rate decay, local optima, etc.; the debugging neural networks section includes hyperparameter tuning, sampling scales, hyperparameter tuning experience, BN, softmax, deep learning framework selection, TensorFlow installation and use, etc.
-
Practical Experience: including decision-making, orthogonalization, F1 score, validation sets and test sets, dataset acquisition and splitting, Bayes error, the relationship between AI capability and human capability, improvingThe general process of AI systems, bias in datasets, manual analysis errors, label correction, transfer learning, multi-task learning, end-to-end learning, dataset creation, etc.
-
Intelligent Vision: including convolutional neural networks, convolution operations, edge detection, padding, convolution stride, 3D convolution, multiple filters, convolutional layers, pooling layers, complete convolutional networks, LeNet-5, AlexNet, VGG, residual networks, 1×1 convolution, Inception networks, object detection, object localization, keypoint detection, sliding window detection, YOLO detection, non-max suppression, style transfer, discrepancy verification, neural style transfer networks, content loss functions, as well as practical programming for building convolutional neural networks, residual networks, vehicle detection for autonomous driving, style transfer, face recognition, etc.
-
Intelligent Speech: including recurrent sequence models, sequence model datasets, RNN, RNN computation process and backpropagation, LSTM, GRU, bidirectional recurrent neural networks, deep RNN, AI writing, intelligent music, intelligent composition, natural language processing and word embeddings, word2vector, negative sampling, GloVe model, sentiment classification, AI bias, seq2seq, Beam search, attention models, speech recognition, wake word detection, machine translation, and other practical programming.
-
Large Models: including the origin of large models, why they are "large" models, Transformer, self-attention mechanism, pre-training + fine-tuning, Tokenizer, unsupervised learning, masked language model (MLM), autoregressive generation, computing power, instruction fine-tuning, RLHF, alignment, text generation, understanding and reasoning, tool use, hallucination generation, logical confusion, bias and discrimination, pitfall avoidance guide, local deployment, prompt engineering, model compression and quantization, fun projects, lightweight revolution, customization era, multi-agent, new model architectures.
-
Agent: including what an Agent is, application domains, the evolution of agents, core traits, technical boundaries, development history, frameworks and platforms, perception module, decision-making module, execution module, memory module, tool invocation module, collaboration module, programming language fundamentals, interfaces and networks, data processing fundamentals, scripting programming, the brain of large model LLM Agents, Prompt engineering, limitations, vector databases, chroma, developing the first Agent, tool invocation, multi-Agent collaboration, performance optimization, deployment and launch, Docker containers, interface encapsulation, monitoring and logging, integration with existing systems, development framework LangChain, AutoGPT, domestic frameworks, and framework selection.
-
RAG: including the origin of RAG, core logic, standard RAG workflow, overall architecture, vector databases, principles of vector retrieval, document processing, text cleaning, chunking (Chunk), metadata, sparse retrieval and dense retrieval, hybrid retrieval, reranking (Rerank), retrieval performance evaluation, generation module, context window, avoiding hallucinations, minimal starting point, local knowledge base question answering, data engineering, multi-format document processing, knowledge base construction process, incremental updates, advanced tuning, retrieval optimization, chunking tuning, vector model selection, prompt engineering, fine-tuning ideas, evaluation and iteration, automatic evaluation metrics, manual evaluation plans, error analysis, Advanced RAG, Modular RAG, adaptive RAG, multi-hop retrieval, RAG+fine-tuning, RAG+attention mechanism, RAG+multimodal, long-text RAG, multimodal Embedding, multimodal retrieval and generation, speed optimization, distributed deployment, cost control, monitoring and operations, enterprise internal knowledge base question answering, customer service intelligent question answering system, academic literature retrieval and summarization, legal/medical domain-specific RAG, RAG pitfalls, RAG+Agent, on-device RAG, RAG without vector database.
-
Multimodal AI programming: including what multimodal AI is, modalities, a brief history of multimodal AI, environment setup, cross-modal alignment, feature fusion, interaction paradigms, CLIP, BLIP series, FLAVA, image-text generation, visual question answering VQA, image-text annotation and datasets, text-to-audio/audio-to-text, video multimodality, 3D multimodality, multimodal fusion in practice, large models+multimodal, open-source multimodal large model deployment, instruction fine-tuning, LoRA, data processing, loss functions, training strategies, evaluation metrics, common Bug troubleshooting, model compression, API service setup, frontend integration, mobile/edge deployment, compliance and copyright, high-frequency interview points.
-
AI engineering: including what AI engineering is, deep learning≠AI engineering, core standards of industrial-grade AI, the four pillars of engineering, engineering differences, engineering jargon, data governance, the three major killers of launch, industrial-grade data pipeline, data drift, data version management, data pipeline, model slimming, slow model inference, the four major slimming techniques, optimization priorities, compilation and acceleration, framework selection, acceleration selection, one-click acceleration process, deployment introduction, model encapsulation, containers and orchestration, Docker containers, Kubernetes, cloud deployment and edge deployment, monitoring, operations, model hot updates, rollback mechanisms, continuous iteration, MLOps.
-
AI Financial Quantitative Trading: including the essence of quantitative trading, AI quantitative trading, industry jargon in finance, financial data, what AI can and cannot do in quantitative trading, AI quantitative trading workflow, financial data classification, open-source data, data cleaning, feature engineering, standardization/normalization, time series data processing, dataset splitting, data augmentation, MLP, CNN, model evaluation, financial time series, gating mechanisms, Transformer, self-attention mechanism, multi-factor stock selection, GNN, linked stock selection strategies, strategy design, signal generation, position management, take-profit and stop-loss, backtesting, backtesting framework selection, survivor bias, multi-model fusion, risk control, paper trading testing, live trading interface integration, live trading monitoring, live trading optimization, overfitting, data leakage, transaction costs.
-
Neuromorphic SNN: including why SNN is needed, deep learning and the brain, three major pain points, spiking neural networks, neurons, synapses, spikes, thresholds, membrane potential, refractory period, rate coding, temporal coding, FC-SNN, SCNN, normalized SNN, comparison with RNN/Transformer, STDP, approximate gradients, ANN to SNN conversion, weight conversion, one-click CNN to SNN, handwritten digit recognition SNN, neuromorphic chips, event camera + SNN, low-power edge AI, industry status.
-
Reinforcement Learning: including what reinforcement learning is, the four major components of RL, MDP, Bellman equation, from tables to neural networks, Q table, DQN, Double/Dueling/Noisy DQN, reward design, policy gradient, REINFORCE, Actor-Critic, A2C, PPO, continuous action space, DDPG, TD3, SAC, sparse rewards, auxiliary rewards, training techniques, RLHF, large model alignment, DPO.
The website introduction states that after graduating from the website, finding a job in artificial intelligence is not a problem, and those with talent can develop various economically valuable artificial intelligence programs. The author hopes to help more Chinese people master artificial intelligence technology and believes that artificial intelligence is an opportunity for China.
Regarding facts such as fees, prices, platform, specific study duration, and certificates, the materials do not currently provide them.
