Similarity Metric Learning Task

A Similarity Metric Learning Task is a machine learning task where the goal is to learn a similarity scoring model (tha scores the similarity between two or more objects).

• Context:
  • Input: paired or grouped examples (to capture relative similarities).
  • ouput a similarity scoring model (that produces a similarity score - rather than]]class probabiliti]]es).
  • It can rely less on independently labeled examples like in classification or regression.
  • It can solved by a Similarlity Metric Learning System (that implements a similarity metric learning algorithm, such as via siamese neural networks and contrastive learning).
  • It can be applied to Recommendation Tasks, Information Retrieval, Face Recognition, etc.
  • ...
• Example(s):
  • A Face Similarity Metric Learning Task to learn a face similarity scoring model based on face image pairs.
  • A Word Similarity Metric Learning Task to learn a word embedding model based on word co-occurrence patterns.
  • A Text Chunk Similarity Metric Learning Task to learn a text chunk embedding model based on similar and dissimilar text chunks.
  • A Question Similarity Metric Learning Task to learn a question similarity model based on duplicate and non-duplicate question pairs.
  • A Product Similarity Metric Learning Task to learn a product similarity model based on groups of commonly bought items.
  • A Resume Similarity Metric Learning Task to learn a resume similarity scoring model to match resumes to job postings based on similarities.
  • An Anomaly Detection Similarity Metric Learning Task to learn an anomaly scoring model to detect abnormal network activity based on normal traffic examples.
  • A Document Similarity Metric Learning Task to learn a document similarity model based on relevant and non-relevant document pairs.
  • A Fraud Detection Similarity Metric Learning Task to learn a fraud scoring model based on examples of legitimate and fraudulent transactions.
  • ...
• Counter-Example(s):
  • A Classification Task that assigns category labels to examples.
  • A Regression Task that predicts a numeric target value.
  • A Clustering Task that groups unlabeled examples.
  • ...
• See: Siamese Neural Network, Similarity Function, Distance Metric Learning.

---

---

References

2021

• (Liu et al., 2021) ⇒ Xiaoqian Liu, Xiaoyu Tang, and Shuying Chen. (2021). “Learning a Similarity Metric Discriminatively with Application to Ancient Character Recognition." In: 14th International Conference, KSEM 2021, Tokyo, Japan, September 13–15, 2021, Proceedings. Springer.
  • QUOTE: "The process of learning good representation in deep learning may prove difficult when the data is insufficient. In this paper, we propose a Siamese similarity network for one-shot ..."
  • NOTE: It explores similarity learning for recognizing ancient characters using limited data.

2011

• (Wang et al., 2011) ⇒ Tinghuai Wang, Shengjin Wang, and Xinyu Ding. (2011). “Learning a Similarity Metric Discriminatively for Pose Exemplar Based Action Recognition." In: 2011 4th International Congress on Image and Signal Processing. IEEE.
  • QUOTE: "Exemplar-based action recognition has the advantages of being compact and time-invariant. But how to select suitable exemplars and measure the pose similarities between frames ..."
  • NOTE: It applies similarity learning techniques for exemplar-based action recognition.

2010

• (Nguyen & Bai, 2010) ⇒ Huu Viet Nguyen, and Li Bai. (2010). “Cosine Similarity Metric Learning for Face Verification." In: Asian Conference on Computer Vision. Springer.
  • QUOTE: "... In this paper we propose a new method, named the Cosine Similarity Metric Learning (CSML) for learning a distance metric for facial verification. The use of cosine similarity in our ..."
  • NOTE: It focuses on cosine similarity metric learning for face verification.

2005

• (Chopra et al., 2005) ⇒ Sumit Chopra, Raia Hadsell, and Yann LeCun. (2005). “Learning a Similarity Metric Discriminatively, with Application to Face Verification". In: CVPR (1). pp. 539-546.
  • QUOTE: "In this paper we address the task of learning a similarity function from data. ... Our experiments demonstrate that the proposed techniques can learn a similarity metric ..."
  • NOTE: It introduces techniques for similarity learning focusing on applications to face verification.

1995

• (Lowe, 1995) ⇒ David G. Lowe. (1995). “Similarity Metric Learning for a Variable-Kernel Classifier." In: Neural Computation.
  • QUOTE: "Nearest-neighbor interpolation algorithms have many useful properties for applications to learning, but they often exhibit poor generalization. In this paper, it is shown that much better ..."
  • NOTE: It examines similarity metric learning to improve nearest-neighbor based classification.

---