FaceNet: Face Recognition and Clustering

11 Apr 2020

A Unified Embedding for Face Recognition and Clustering

Paper: FaceNet: A Unified Embedding for Face Recognition and Clustering
Authors: Florian Schroff (Google), Dimitry Kalenichenko (Google) and James Philbin (Google)
Area: Computer Vision, Clustering, Classification, Deep Learning
Year: 2015
Highlighted Paper

Background:

Euclidean space, distance and vector norms

Nearest neighbor/k-means/clustering

CNNs: Stanford's CS231N course notes, Deep learning book chapter 9

Important architectures for this paper:

Key Contributions:

Learns a mapping from face images to a compact Euclidean space where distances directly correspond to measure of face similarity. The method uses a CNN to directly optimize the embedding itself.
To train - use triplets of roughly aligned matching/non-matching face patches generated using an online triplet mining method.

Why is this novel? Unlike previous approaches where a final classification layer is used to predict the class, here we are leaning an embedding which can then be used for various classification purposes.

Model Triplet Loss:
$$\sum_{i}^{N}[\lVert f(x_{i}^{a}) - f(x_{i}^{p}) \rVert_{2}^{2} - \lVert f(x_{i}^{a}) - f(x_{i}^{n}) \rVert_{2}^{2} + \alpha)]$$ $a$ is the anchor, $p$ is a positive and $n$ is a negative. You essentially want to optimize for making the positive closer to the anchor than the negative.

Triplet Selection:
We focus on the online generation and use large mini-batches in the order of a few thousand exemplars and only compute the $argmin$ and $argmax$ within that mini-batch.

CNN Architecturse:
Existing architectures - Inception model and the Zeiler&Fergus model.

Some implementations:
tbmoon's implementation using PyTorch
timseler's implementation

Datasets
Labeled Faces in the Wild (LFW) courtsey of UMass
YouTube Faces DB

Sequence to Sequence Learning with Neural Networks

25 Oct 2019

Using LSTMs for general sequence-to-sequence problems

Deep neural networks are powerful machine learning models that achieve excellent performance on difficult problems. If there exists a parameter setting of a large DNN that achieves good results, supervised backpropagation will find these parameters and solve the problem.

The Problem

Many important problems are best expressed with sequences whose lengths are not known a-priori. For example, speech recognition and machine translation are sequential problems. Likewise, question-answering can also be seen as mapping a sequence of words representing the question to a sequence of words representing the answer.

The Approach

The core idea is to use one LSTM to read the input sequence, one timestep at a time, to obtain a large fixed-dimensional vector representation, and then use another LSTM to extract the output sequence from that vector.

The second LSTM is essentially a recurrent neural network language model, except that it is conditioned on the input sequence.

Key Properties

The LSTM learns to map an input sequence of variable length into a fixed-dimensional vector representation
The encoder-decoder architecture separates input processing from output generation
Attention mechanisms (in later work) address limitations of fixed-size representations

Related Work

Kalchbrenner and Blunsom, "Recurrent continuous translation models" (EMNLP, 2013)
Cho et al., "Learning phrase representations using RNN encoder-decoder for statistical machine translation" (2014)

Beam Search Algorithm

15 May 2019

A heuristic search algorithm for finding optimal sequences

Overview

Beam search is a heuristic search algorithm commonly used in sequence-to-sequence models for tasks like machine translation, speech recognition, and text generation. It explores a graph by expanding the most promising nodes while keeping only a limited number of candidates at each level.

How It Works

Beam search uses breadth-first search to build its search tree. At each level:

Generate all successors of the states at the current level
Sort them in increasing order of heuristic cost
Keep only the top $\beta$ states (the beam width)
Expand only those states in the next iteration

The greater the beam width, the fewer states are pruned. With $\beta = 1$, beam search becomes greedy search. With $\beta = \infty$, it becomes exhaustive breadth-first search.

Why Use Beam Search?

Beam search maintains tractability in large systems with insufficient memory to store the entire search tree. It provides a balance between:

Greedy search: Fast but may miss optimal solutions
Exhaustive search: Optimal but computationally infeasible

Applications

Neural Machine Translation: Finding the most likely translation
Speech Recognition: Decoding audio to text
Image Captioning: Generating descriptions for images
Text Generation: Producing coherent sequences in LLMs

Limitations

Beam search is not guaranteed to find the optimal solution since it prunes the search space. The choice of beam width $\beta$ is a trade-off between search quality and computational cost.

Rishabh Ranawat