The Multiverse Loss for Robust Transfer Learning

By Etai Littwin, Lior Wolf. 2015

24 citations recorded by the source corpus.

Deep learning techniques are renowned for supporting effective transfer learning. However, as we demonstrate, the transferred representations support only a few modes of separation and much of its dimensionality is unutilized. In this work, we suggest to learn, in the source domain, multiple orthogonal classifiers. We prove that this leads to a reduced rank representation, which, however, supports more discriminative directions. Interestingly, the softmax probabilities produced by the multiple classifiers are likely to be identical. Experimental results, on CIFAR-100 and LFW, further demonstrate the effectiveness of our method.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Ocean Engineering, Representation & Self-Supervised Learning, Meta & Continual Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Ali Farhadi
2. Mohammad Rastegari
3. Bharath Hariharan
4. Abhinav Kumar Gupta
5. Mohit Bansal
6. Jianfeng Gao
7. Dacheng Tao
8. Trevor Darrell

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Xiaogang Wang
3. Dacheng Tao
4. Jitendra Malik
5. Trevor Darrell
6. Andrea Vedaldi
7. Ming-Hsuan Yang
8. Chunhua Shen

In Ocean Engineering

1. Joseph Keshet
2. Bharath Hariharan
3. Yossi Adi
4. Ross B. Girshick
5. Gabi Shalev

In Representation & Self-Supervised Learning

1. Ross B. Girshick
2. Bharath Hariharan
3. Xiaoming Liu
4. Xi Yin
5. Ali Farhadi
6. Joseph Keshet
7. Mohammad Rastegari
8. Hessam Bagherinezhad

In Meta & Continual Learning

1. Ali Farhadi
2. Joseph Keshet
3. Bharath Hariharan
4. Mohammad Rastegari
5. Yossi Adi
6. Ross B. Girshick
7. Gabi Shalev

Citations in the corpus

• Low-shot visual object recognition (2016, background): “As a final baseline, we also consider the multiverse loss [26] which was shown to improve transfer learning performance.”
• Multi-Task Convolutional Neural Network for Pose-Invariant Face Recognition (2017, background): “Recent CNN-based approaches mainly focus on exploring the effects of 3D model-based face alignment [50], larger datasets [42], [50], or new metric learning algorithms [33], [42], [46], [54] on face recognition performance.”
• Out-of-Distribution Detection using Multiple Semantic Label Representations (2018, methodology): “In [26] the authors presented a novel technique for robust transfer learning, where they proposed to optimize multiple orthogonal predictors while using a shared representation.”
• LCNN: Lookup-Based Convolutional Neural Network (2016, background): “Few-Shot Learning: The problem of learning novel categories has been studied in [40, 2, 30].”
• Coupled End-to-End Transfer Learning with Generalized Fisher Information (2018, unknown): “More recently, in [25], several factors (including width, depth, density, etc.”
• Low-Shot Visual Recognition by Shrinking and Hallucinating Features (2016, background): “There have also been novel losses proposed explicitly to aid transfer, such as the multiverse loss of Littwin and Wolf [29].”