---

[Github] https://github.com/createmomo (You can also find the Wechat Public Account here)

Most blog posts have both English and Chinese versions, with only a few exceptions written solely in Chinese.

Collections (Post Series)

• [2025-2026, Planning…]
  • Language Model Learning & Practice (Student-Oriented Series) (TBD)
  • Current Trends in Goal-Guided Conversational AI Models (TBD)
  • Using Language Models in Specific Domains (TBD)
• [2023-2025] Past of Goal-Guided Conversational AI Models
  • 1 Introduction
  • 2 Target-Guided Open-Domain Conversation (ACL2019)
  • 3 Proactive Human Machine Conversation with Explicit Conversation Goal (ACL2019)
  • 4 Towards Conversational Recommendation Over Multi-Type Dialogs (ACL2020)
  • 5 Knowledge Graph Grounded Goal Planning for Open-Domain Conversation Generation (AAAI2020)
  • 6 Towards Topic Guided Conversational Recommender System (Arxiv 2020)
  • 7 Towards Effective Automatic Debt Collection with Persona Awareness (EMNLP2023)
  • 8 Reinforcement Learning of Cooperative Persuasive Dialogue Policies Using Framing (COLING 2014)
  • 9 Dialogue Scenario Collection of Persuasive Dialogue with Emotional Expressions via Crowdsourcing (LREC 2018)
  • 10 Persuasion for Good: Towards a Personalized Persuasive Dialogue System for Social Good (ACL2019)
  • 11 Quick Review
  • 12 OTTers: One-turn Topic Transitions for Open-Domain Dialogue (ACL&IJCNLP 2021)
  • 13 Towards a Universal NLG for Dialogue Systems and Simulators with Future Bridging（2021）
  • 14 SalesBot: Transitioning from Chit-Chat to Task-Oriented Dialogues (ACL, 2022)
• [2024-2025] Use AI to Detect AI-Generated Text
  • 1 Introduction
  • 2 Testbed1~2
  • 3 Testbed3~4
  • 4 Testbed5~8
  • 5 Key Experimental Details
  • 6 Results, ChatGPT&Human
  • 7 Results, Testbed1
  • 8 Results, Testbed2~4
  • 9 Results, Testbed5
  • 10 Results, Testbed6 & Tips for Setting Classification Thresholds
  • 11 Results, Testbed7
  • 12 Results, Testbed8，Attempt to Evade being Detected by AI
  • 13 Some interesting analyses
  • 14 Ghostbuster
  • 15 Ghostbuster (Results)
  • 16 Ghostbuster (Make the Detector More Robust)
  • 17 Ghostbuster (Again How to Avoid Being Detected by AI)
• [2023-2024] Using Language Models in Specific Domains
  • 1 Introduction
  • 2 Domain-specific Training Data
    • [Any Domain] Use Unlabelled Text to Improve Instruction Following Language Models (Notes 1, 2, 3, 4, 5)
    • [Medical/Health] ChatDoctor (Notes 1, 2, 3; Slides 1, 2, 3)
    • [Medical/Health] MedicalGPT-zh (Notes, Slides)
    • [Medical/Health] MING (Notes, Slides)
    • [Medical/Health] SoulChat (Notes, Slides)
    • [Mobile Interaction] Tiny Models, Mighty Powers - ReALM (1, 2, 3, 4)
  • 3 Automatic Model Evaluation
    • [Any Domain] Evaluating Language Models with Language Models (1 Introduction)
    • [Any Domain] Evaluating Language Models with Language Models (2 PandaLM)
    • [Any Domain] Evaluating Language Models with Language Models (3 Shepherd, 1,2,3,4)
    • [Medical/Health] Comparing ChatDoctor and ChatGPT3.5 using BERT-Score
• [2024] Tiny Models, Mighty Powers - ReALM (1, 2, 3, 4)
• [2023-2024] Use Unlabelled Text to Improve Instruction Following Language Models
  • 1 Introduction
  • 2 Initialisation & Self-Augmentation
  • 3 Self-Curation & Experiments
  • 4 Some Conclusions I
  • 5 Some Conclusions II
• [2023] Evaluating Language Models with Language Models
  • 1 Introduction
  • 2 PandaLM
  • 3 Shepherd (1, 2, 3, 4)
• [2022-2024] Conference “Interesting”s
  • Interesting · LREC-COLING&NAACL2024
  • Interesting · ACL2023
  • Interesting · EACL2023
  • Interesting · EMNLP2023
  • Interesting · ACL2022 (Findings)
  • Interesting · ACL2022 (Short Papers)
  • Interesting · ACL2022 (Long Papers)
  • Interesting · NeurIPS2022
  • Interesting · EMNLP2022 (Findings&CL Papers)
  • Interesting · EMNLP2022 (Main Conference)
• [2023] Chinese Natural Language Understanding, NLU, in Dialogue Systems
  • 1&2 Task Introduction&Chinese vs English in NLP
  • 3.1 Academic Methods （Joint Training）
  • 3.2 Academic Methods （Separate Training）
  • 4 Industry Methods
  • 5&6 Chinese Dialogue System Challenge Track & Resources
• [2021] Fantastic Trees (Decision Trees, Random Forest, Adaboost, Gradient Boosting DT, XGBoost)
• [2020] Improving Your English Communication Skills (Writing Emails, Speaking English and Building ePortfolio)
• [2017] CRF Layer on the Top of BiLSTM
  • CRF Layer on the Top of BiLSTM - 1 Outline and Introduction
  • CRF Layer on the Top of BiLSTM - 2 CRF Layer (Emission and Transition Score)
  • CRF Layer on the Top of BiLSTM - 3 CRF Loss Function
  • CRF Layer on the Top of BiLSTM - 4 Real Path Score
  • CRF Layer on the Top of BiLSTM - 5 The Total Score of All the Paths
  • CRF Layer on the Top of BiLSTM - 6 Infer the Labels for a New Sentence
  • CRF Layer on the Top of BiLSTM - 7 Chainer Implementation Warm Up
  • CRF Layer on the Top of BiLSTM - 8 Demo Code
    

Notes (Single Post)

• [2023-2025] Open-Source Language Model Pocket
• [2023] Using ColossalAI SFT in Kaggle or Colab (in Chinese)
• [2023] General Understanding of Decoding Strategies Commonly Used in Text Generation
• [2022] Understand Gradient Checkpoint in Pytorch
• [2022] Baidu World Conference 2022: AI Applications (in Chinese, Notes)
• [2021] GPT Understands, Too
• [2021] Super Git Revision Notes
• [2021] Baidu World Conference 2021: AI Applications (in Chinese, Notes)
• [2019] Probabilistic Graphical Models Revision Notes
• [2018] Super Machine Learning Revision Notes

Paper Explained

• [2021] Few-Shot Text Classification with Distributional Signatures (ICLR 2020) Part1
• [2021] Few-Shot Text Classification with Distributional Signatures (ICLR 2020) Part2
• [2021] Few-Shot Text Classification with Distributional Signatures (ICLR 2020) Part3

---

Detailed Links:
* Fantastic Trees (Decision Trees, Random Forest, Adaboost, Gradient Boosting DT, XGBoost)

* Probabilistic Graphical Models Revision Notes

• Representations
  • Bayesian Network (directed graph)
    • Defination
    • Reasoning Patterns in Bayesian Network
    • Flow of Probabilistic Influence (active trial)
    • Independencies
    • d-seperation
    • I-Maps (Indenpendency Map)
    • Factorisation and I-Maps
    • Naive Bayes
    • Template Models
      • Temporal Models (involve over time)
      • 2 Time-Slice Bayesian Network (2TBN)
      • Plate Models
    • Conditional Probability Distribution (CPD)
      • General CPD
      • Table-based CPD
      • Context-specific Independence
      • Tree-Structured CPD
      • Multiplexer CPD
      • Noise OR CPD
      • Sigmoid CPD
      • Continuous Variables
  • Markov Network (undirected graph)
    • Markov Network Fundamentals
      • Pairwise Markov Networks
      • General Gibbs Distribution (a more general expression)
      • Induced Markov Network (connects every pair of nodes that are in the same factor)
      • Factorization
      • Conditional Random Fields
      • Independencies in Markov Networks
    • Local Structure in Markov Networks
      • Log-linear Models (CRF, Ising Model, Metric MRFs)
    • Decision Making
      • Maximum Expected Utility
      • Utility Functions
      • Value of Perfect Information
    • Knowledge Engineering
      • Generative vs. Discriminative
      • Designing a graphical model (variable types)
      • Structure
      • Parameters: Local Structure
      • Iterative Refinement
• Inference
  • Conditional Probability Queries (Overview)
    • Sum-Product in Bayes Network and Markov Network
    • Evidence: Reduced Factors
    • Summary of Sum-Product Algorithm
  • MAP (Maximum A Posterior) Inference (Overview)
    • Max-Product
  • Variable Elimination
    • Variable Elimination Algorithm
      • Elimination in Chains
      • Elimination in a more complicated BN
      • Variable Elimination with Evidence
      • Variable Elimination in MNs
      • Summary Variable Elimination Algorithm
    • Complexity of Variable Elimination
    • Graph-based Perspective on Variable Elimination
    • Finding Elimination Orderings
    • Variable Elimination (Summary)
• Learning
  

* Super Machine Learning Revision Notes

• Activation Functions
• Gradient Descent
  • Computation Graph
  • Backpropagation
  • Gradients for L2 Regularization (weight decay)
  • Vanishing/Exploding Gradients
  • Mini-Batch Gradient Descent
  • Stochastic Gradient Descent
  • Choosing Mini-Batch Size
  • Gradient Descent with Momentum (always faster than SGD)
  • Gradient Descent with RMSprop
  • Adam (put Momentum and RMSprop together)
  • Learning Rate Decay Methods
  • Batch Normalization
• Parameters
  • Learnable and Hyper Parameters
  • Parameters Initialization
  • Hyper Parameter Tuning
• Regularization
  • L2 Regularization (weight decay)
  • L1 Regularization
  • Dropout (inverted dropout)
  • Early Stopping
• Models
  • Logistic Regression
  • Multi-Class Classification (Softmax Regression)
  • Transfer Learning
  • Multi-task Learning
  • Convolutional Neural Network (CNN)
    • Filter/Kernel
    • Stride
    • Padding (valid and same convolutions)
    • A Convolutional Layer
    • 1*1 Convolution
    • Pooling Layer (Max and Average Pooling)
    • LeNet-5
    • AlexNet
    • VGG-16
    • ResNet (More Advanced and Powerful)
    • Inception Network
    • Object Detection
      • Classification with Localisation
      • Landmark Detection
      • Sliding Windows Detection Algorithm
      • Region Proposal (R-CNN)
      • YOLO Algorithm
        • Bounding Box Predictions (Basics of YOLO)
        • Intersection Over Union
        • Non-max Suppression
        • Anchor Boxes
    • Face Verification
      • One-Shot Learning (Learning a “similarity” function)
        • Siamese Network
        • Triplet Loss
      • Face Recognition/Verification and Binary Classification
    • Neural Style Transfer
  • Sequence Models
    • Recurrent Neural Network Model
    • Gated Recurrent Unit (GRU)
      • GRU (Simplified)
      • GRU (Full)
    • Long Short Term Memory (LSTM)
    • Bidirectional RNN
    • Deep RNN Example
    • Word Embedding
      • One-Hot
      • Embedding Matrix ($E$)
      • Learning Word Embedding
      • Word2Vec & Skip-gram
      • Negative Sampling
      • GloVe Vector
      • Deep Contextualized Word Representations (ELMo, Embeddings from Language Models)
    • Sequence to Sequence Model Example: Translation
      • Pick the most likely sentence (Beam Search)
        • Beam Search
        • Length Normalisation
        • Error Analysis in Beam Search (heuristic search algorithm)
      • Bleu Score
      • Combined Bleu
      • Attention Model
  • Transformer (Attention Is All You Need)
  • Bidirectional Encoder Representations from Transformers (BERT)
• Practical Tips
  • Train/Dev/Test Dataset
  • Over/UnderFitting, Bias/Variance, Comparing to Human-Level Performance, Solutions
  • Mismatched Data Distribution
  • Input Normalization
  • Use a Single Number Model Evaluation Metric
  • Error Analysis (Prioritize Next Steps)

[2017] CRF Layer on the Top of BiLSTM (BiLSTM-CRF)

• CRF Layer on the Top of BiLSTM - 1 Outline and Introduction
• CRF Layer on the Top of BiLSTM - 2 CRF Layer (Emission and Transition Score)
• CRF Layer on the Top of BiLSTM - 3 CRF Loss Function
• CRF Layer on the Top of BiLSTM - 4 Real Path Score
• CRF Layer on the Top of BiLSTM - 5 The Total Score of All the Paths
• CRF Layer on the Top of BiLSTM - 6 Infer the Labels for a New Sentence
• CRF Layer on the Top of BiLSTM - 7 Chainer Implementation Warm Up
• CRF Layer on the Top of BiLSTM - 8 Demo Code

---

---

This note summarises the Youtube Videos published by Josh Starmer (Youtube Account: StatQuest with Josh Starmer). I would like to say a big thank you to him and his super useful videos!

Read More

---

This page lists the notes of interesting papers at different research topics of Natural Language Processing (NLP). Each note briefly describes the main points of each paper. Hope this would be helpful for you to quickly get the ideas of them. (Please be free to correct me if you found mistakes.)

Read More

---

This is the note of this online course, Improving Your English Communication Skills on Coursera.

Read More

[Last Updated: 2020.02.23]

This note summarises the online course, Probabilistic Graphical Models Specialization on Coursera.
Any comments and suggestions are most welcome!

Read More

[Last Updated: 06/01/2019]

This article aims to summarise:

• basic concepts in machine learning (e.g. gradient descent, back propagation etc.)
• different algorithms and various popular models
• some practical tips and examples were learned from my own practice and some online courses such as Deep Learning AI.

If you a student who is studying machine learning, hope this article could help you to shorten your revision time and bring you useful inspiration. If you are not a student, hope this article would be helpful when you cannot recall some models or algorithms.

Moreover, you can also treat it as a “Quick Check Guide”. Please be free to use Ctrl+F to search any key words interested you.

Any comments and suggestions are most welcome!

Read More

---

• Reading for Learning English
  • [2017.07.25 -] Successful Writing at Work, Ninth Edition
  • [2017.09.27 -] The Curse of the Cheese Pyramid (Geronimo Stilton)
  • [2016.12.17 - 2017.09.26] Lost Treasure of the Emerald Eye (Geronimo Stilton)
• Books
  • [2016.05.01 -] Love in the Time of Cholera
  • [2015.02.24 - 2016.12.24] 1Q84
  • [2013.12.07 - 2015.04.10] The Three-Body Problem Series
• TV Series & Movies
  • [2018.03.02 - ] The Big Bang Theory
  • [2018.01.27 - 2018.03.01] Sherlock
  • [2016.04.13 - 2018.01.26] Downton Abbey
  • [2015.04.01 - 2017.05.26] the Vampire Diaries
• Piano
  • [2016.09.01 - 2017.02.05] OST : Just One Time is Enough (Aska Yang)
• Sports (First Time)
  • [2017] Skiing
  • [2014] Ice Skating
  • [2007] Swimming

3.4 Demo

In this section, we will make two fake sentences which only have 2 words and 1 word respectively. Moreover, we will also randomly generate their true answers. Finally, we will show how to train the CRF Layer by using Chainer v2.0. All the codes including the CRF layer are avaialbe from GitHub.

Read More

3 Chainer Implementation

In this section, the structure of code will be explained. In addition, an important tip of implementing the CRF loss layer will also be given. Finally, the Chainer (version 2.0) implementation source code will be released in the next article.

Read More

2.6 Infer the labels for a new sentence

In the previous sections, we learned the structure of BiLSTM-CRF model and the details of CRF loss function. You can implement your own BiLSTM-CRF model by various opensource frameworks (Keras, Chainer, TensorFlow etc.). One of the greatest things is the backpropagation of on your model is automatically computed on these frameworks, therefore you do not need to implement the backpropagation by yourself to train your model (i.e. compute the gradients and to update parameters). Moreover, some frameworks have already implemented the CRF layer, so combining a CRF layer with your own model would be very easy by only adding about one line code.

In this section, we will explore how to infer the labels for a sentence during the test when our model is ready.

Read More