Mate&Few-Shot Learning

Mate&Few-Shot Learning

1. 基本概念

1. Meta Learning
   • $Few-shotlearning$ is a kind of $metalearning$.
   • Meta learning: $learntolearn$.
2. Supervised Learning vs. Few-Shot Learning
   1. Traditional supervised learning:
      • $Test$ samples are $neverseenbefore$.
      • $Test$ samples are from $knownclasses$.
   2. Few-shot learning:
      • $Query$ samples are $neverseenbefore$.
      • $Query$ samples are from $unknownclasses$.
3. 术语
   • Training Set：
   • Support Set：
     • $k$-way: the support set has $k$ classes.
     • $n$-shot: every class has $n$ samples.

       • 3-way is easier than 6-way;
       • 2-shot is easier than 1-shot.

   • Query：
4. Idea: Learn a Similarity Function
   • Basic Idea:
     • Learn a similarity function: $sim(x,x^{\ast })$.
     • Ideally, $sim(x_1,x_2)=1$, $sim(x_1,x_3)=0$, and $sim(x_2,x_3)=0$.
   • Step:
     • First, learn a similarity function from large-scale training dataset.
     • Then, apply the similarity function for prediction.
       • Compare the $query$ with every sample in the $supportset$.
       • Find the sample with the highest similarity score.
5. Datasets
   • Omniglot
     • Official website: https://github.com/brendenlake/omniglot/
     • TensorFlow: https://www.tensorflow.org/datasets/catalog/omniglot

---

2. Siamese Network

2.1 Learning Pairwise Similarity Scores

Ref:

• Bromley et al. Signature verification using a Siamese time delay neural network. In NIPS. 1994.
• Koch, Zemel, & Salakhutdinov. Siamese neural networks for one-shot image recognition. In ICML, 2015.

1. Data for Training set
   Select 2 pieces of training data each time and label them.
   
2. CNN for Feature Extraction
   
3. Training Siamese Network
   • Forword Network
   • BackWord Network
     Update the parameter of CNN.
     
4. One-shot Prediction
   The training data (for the Siamese network) does not contain the support set classes and the query.
   

2.2 Triplet Loss

Ref:

• Schroff, Kalenichenko, & Philbin. Facenet: A unified embedding for face recognition and clustering. InCVPR, 2015

1. Data for Training set
   Select 3 pieces of training data each time and label them.
   
2. CNN for Feature Extraction
   Feature Extraction use the same CNN.
   
3. Triplet Loss
   
4. One-Shot Prediction
   

2.3 Basic Idea of Few-Shot Learning

• Train a $Siamesenetwork$ on large-scale training set.
• Given a $supportset$ of 𝑘-way 𝑛-shot.
  • 𝑘-way means 𝑘 classes.
  • 𝑛-shot means every class has 𝑛 samples.
  • The training set does not contain the 𝑘 classes.
• Given a $query$, predict its class.
  • Use the Siamese network to compute similarity or distance.

---

3. Pretraining and Fine Tuning

1. Cosine Similarity
   

2. Softmax Function
   

3. Softmax Classifier(全连接层+softmax函数)
   Here, 𝑘 is number of classes, and 𝑑 is number of features.
   

3.1 Few-Shot Prediction Using Pretrained CNN

Reference:

• Dhillon, Chaudhari, Ravichandran, & Soatto. A baseline for few-shot image classification. In ICLR, 2020.
• Chen, Wang, Liu, Xu, & Darrell. A New Meta-Baseline for Few-Shot Learning. arXiv, 2020

1. Pretraining
   • Pretrain a CNN for $featureextraction$ (aka embedding).
   • The CNN can be pretrained using $standardsupervisedlearning$ or $Siamesenetwork$.
     
2. Deal with the Support set
   
3. Making Few-Shot Prediction
   $q$ is Query.
   
4. Summary
   

---

3.2 Benefit of Fine Tuning

Reference:

• Chen, Liu, Kira, Wang, & Huang. A Closer Look at Few-shot Classification. In ICLR, 2019.
• Dhillon, Chaudhari, Ravichandran, & Soatto. A baseline for few-shot image classification. In ICLR, 2020.
• Chen, Wang, Liu, Xu, & Darrell. A New Meta-Baseline for Few-Shot Learning. arXiv, 2020.

• Fine-Tuning is a improved algorithm for Few-Shot Prediction Using Pretrained CNN.
• The process of Few-Shot Prediction Using Pretrained CNN is:
  
  the $j$ of $x_j$ is the $query$. $W$ and $b$ are from the support set.

1. Trick 1: A Good Initialization
   
   We can train $W$ and $b$ on the support set. (Fine tuning.)
2. Trick 2: Entropy Regularization
   

   

3. Trick 3: Cosine Similarity + Softmax Classifier
   
   

• summary
  

参考：

1. 小样本学习和元学习（中文课程） - Shusen Wang