SlideShare ist ein Scribd-Unternehmen logo
1 von 14
Downloaden Sie, um offline zu lesen
MARS: Motion-Augmented RGB Stream for Action
Recognition (CVPR 2019)
MARS: Motion-Augmented RGB Stream for Action
Recognition (CVPR 2019)
Influence-Balanced Loss for
Imbalanced Visual Classification
ASRI, Dept. of Electrical and Computer Engineering
Seoul National University
Seulki Park Jongin Lim Younghan Jeon Jin Young Choi
Imbalanced Visual Classification
2
Introduction Proposed Method Experiment Conclusion
Many real-world data often exhibit long-tailed distribution.
✓ The model trained on such imbalanced data tends to overfit the majority classes.
✓ That is, the model performs poorly on minority classes.
Problem Definition:
● Input: Long-tailed (imbalanced) training data & uniform-distributed (balanced) test data.
● Goal: To make a robust model that can generalize well on balanced test data.
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Faces (Zhang et al., 2017) Places (Wang et al., 2017) Species (Van Horn et al., 2018) Actions (Zhang et al., 2019)
* Images by authors.
Previous Methods
Introduction Proposed Method Experiment Conclusion
1. Data-level approach
◩ Directly balance the training data distributions by re-sampling or generating synthetic samples
(Chawla’02, Mullick’ 19, Hulse’ 07).
◩ Under-sampling majority classes can lose some valuable information.
◩ Over-sampling or data generation is susceptible to overfitting to certain repetitive samples.
2. Meta-learning approach
◩ Recent meta-learning methods have shown promising results (Shu’ 19, Liu’20, Ren’20).
◩ However, these methods are difficult to implement in practice.
â–Ș Additional unbiased data are required (Shu’19), Meta-sampler is computationally expensive (Ren’20).
3. Re-weighting approach
◩ Assign different weights to each sample according to its importance.
◩ However, they have focused on only global class-level distribution and assign the same weight to all
samples belonging to the same class (CB Loss [Cui’19], LDAM [Cao’19]).
→ “Not all samples in a dataset play an equal role in determining the model parameters. (Cook, 1982)”
3
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Motivation
Introduction Proposed Method Experiment Conclusion
Q. How can we appropriately re-weight each training samples
while preventing the model from overfitting to majority classes?
- Focal Loss [Lin’17] assigns more weights to hard examples, but most hard examples
belong to majority classes as training progresses.
A. Let’s re-weight samples by their direct influences on making the overfitted model!
→ To measure the influence of a sample, we focus on Influence function (Cook, 1982).
4
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Motivation
Introduction Proposed Method Experiment Conclusion
Key observations:
✓ Influence of the majority class is much greater than that of the minor class!
Key idea:
✓ Down-weight the samples that have large influence on the overfitted decision
boundary to make a smoother decision boundary!
5
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
(a) Comparison of Influences
between balanced and imbalanced dataset.
(b) key concept of our approach.
Proposed Method
Introduction Proposed Method Experiment Conclusion
0. Recap: Influence Functions on DNNs (Koh and Liang, ICML, 2017)
Let 𝑓(đ‘„, 𝜔) a model, 𝐿(𝑩, 𝑓(đ‘„, 𝜔)) a loss for a training point (đ‘„, 𝑩).
By definition, Influence of (đ‘„, 𝑩) on parameters of model (𝜔) is given by:
đ›Ș đ‘„; 𝜔 = âˆ’đ»âˆ’1
∇𝝎𝐿 𝑩, 𝑓 đ‘„, 𝜔 ,
where đ» ≝
1
𝑛
Ïƒđ‘–=1
𝑛
∇𝜔
2 𝐿 𝑩𝑖, 𝑓 đ‘„đ‘–, 𝜔
1. Influence-balanced (IB) weighting factor
From đ›Ș đ‘„; 𝜔 , we design the IB weighting factor as follows:
IB đ‘„; 𝜔 = ∇𝝎𝐿 𝑩, 𝑓 đ‘„, 𝜔 1
6
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
(1)
(2)
Proposed Method
Introduction Proposed Method Experiment Conclusion
2. Influence-balanced (IB) Loss
When using the softmax cross-entropy loss, IB weighting factor can be further simplified:
IB đ‘„; 𝜔 = ∇𝝎𝐿 𝑩, 𝑓 đ‘„, 𝜔 1
= 𝑓 đ‘„, 𝜔 − 𝑩 1 ⋅ ℎ 1
where ℎ is a hidden feature vector.
Finally, the influence-balanced loss is given by
đżđŒđ”(𝑩, 𝑓 đ‘„, 𝜔 ) =
𝐿(𝑩, 𝑓 đ‘„, 𝜔 )
𝑓 đ‘„, 𝜔 − 𝑩 1 ⋅ ℎ 1
→ The proposed influence-balanced term constrains the decision boundary to
not overfit to influential majority samples.
7
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
(3)
(4)
Proposed Method
Introduction Proposed Method Experiment Conclusion
3. Influence-balanced (IB) Class-wise Re-weighting
Finally, we add a class-wise re-weighting term 𝜆𝑘 to the IB-loss in 2. as:
đżđŒđ”(𝜔) =
1
𝑚
෍
đ‘„,𝑩 âˆˆđ·đ‘š
𝜆𝑘
𝐿(𝑩, 𝑓 đ‘„, 𝜔 )
𝑓 đ‘„, 𝜔 − 𝑩 1 ⋅ ℎ 1
where 𝜆𝑘 = đ›Œ
𝑛𝑘
−1
Ïƒđ‘˜â€Č
đŸ
𝑛𝑘â€Č
−1. (𝑛𝑘: the number of samples in the 𝑘-th class)
→ The class-wise re-weighting can further control the influences depending on the classes.
8
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
(5)
Proposed Method
Introduction Proposed Method Experiment Conclusion
4. Influence-balanced Training Scheme
The influence-balanced training process comprises two phases:
1) normal training and 2) fine-tuning for balance.
9
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Experimental Results
Introduction Proposed Method Experiment Conclusion
1. Datasets
◩Synthetic data:
â–ȘCIFAR10/100 (10/100 classes), Tiny ImageNet (200 classes)
â–ȘLong-tailed imbalance: the number of samples of 𝑘-th class is set to 𝑛𝑘𝜇𝑘, (𝜇 ∈ 0,1 ).
â–ȘStep-imbalance: the classes are divided into two groups (majority, minority).
◩ Real-world data:
â–ȘiNaturalist 2018: 437,513 images from 8,142 classes (imbalance factor: 500)
※ imbalance factor: the ratio between the most frequent class and the least frequent class.
2. Baselines
We compare our method with the following cost-sensitive loss methods:
◩CE: uses standard cross-entropy loss.
◩Focal Loss [Lin et al., ICCV17]: down-weights well-classified samples and up-weights hard samples.
◩CB Loss [Cui et al., CVPR19]: re-weights the loss inversely proportional to the effective number of samples.
◩LDAM [Cao et al., NeurIPS19]: regularizes the minority classes to have larger margins.
10
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Comparison with the state-of-the-arts
Introduction Proposed Method Experiment Conclusion
Our method achieves the state-of-the-art results on benchmark datasets with various
imbalance factors.
11
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Classification Accuracy (%) of ResNet-32 on imbalanced CIFAR-10 and CIFAR-100 Classification Accuracy (%) of ResNet-18 on
imbalanced Tiny ImageNet.
Classification Accuracy (%) of ResNet-50 on
iNaturalist2018.
Class-wise classification accuracy
Introduction Proposed Method Experiment Conclusion
Our proposed method shows that the significant performance improvement has resulted
from the minority classes, not from the majority classes!
12
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Conclusion
Introduction Proposed Method Experiment Conclusion
✓Discovered that the existing loss-based loss methods can lead a decision boundary of DNNs to
eventually overfit to the majority classes.
✓Designed a novel influence-balanced loss function to re-weight samples more effectively in a
way to alleviate overfitting of decision boundary.
✓Experimentally demonstrated that IB-loss can improve the generalization performance on
imbalanced data.
✓Our method is easy to be implemented and integrated into existing methods.
13
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Conclusion
Introduction Proposed Method Experiment Conclusion
✓Discovered that the existing loss-based loss methods can lead a decision boundary of DNNs to
eventually overfit to the majority classes.
✓Designed a novel influence-balanced loss function to re-weight samples more effectively in a
way to alleviate overfitting of decision boundary.
✓Experimentally demonstrated that IB-loss can improve the generalization performance on
imbalanced data.
✓Our method is easy to be implemented and integrated into existing methods.
14
Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
Contact: seulki.park@snu.ac.kr
Code: https://github.com/pseulki/IB-Loss

Weitere Àhnliche Inhalte

Was ist angesagt?

Understanding random forests
Understanding random forestsUnderstanding random forests
Understanding random forests
Marc Garcia
 
Research of adversarial example on a deep neural network
Research of adversarial example on a deep neural networkResearch of adversarial example on a deep neural network
Research of adversarial example on a deep neural network
NAVER Engineering
 
Machine Learning and Inductive Inference
Machine Learning and Inductive InferenceMachine Learning and Inductive Inference
Machine Learning and Inductive Inference
butest
 

Was ist angesagt? (20)

1.6.data preprocessing
1.6.data preprocessing1.6.data preprocessing
1.6.data preprocessing
 
äșșć·„çŸ„èƒœ2018 ćŒ·ćŒ–ć­Šçż’ăźćżœç”š
äșșć·„çŸ„èƒœ2018 ćŒ·ćŒ–ć­Šçż’ăźćżœç”šäșșć·„çŸ„èƒœ2018 ćŒ·ćŒ–ć­Šçż’ăźćżœç”š
äșșć·„çŸ„èƒœ2018 ćŒ·ćŒ–ć­Šçż’ăźćżœç”š
 
Uncertainty Estimation in Deep Learning
Uncertainty Estimation in Deep LearningUncertainty Estimation in Deep Learning
Uncertainty Estimation in Deep Learning
 
[DLèŒȘèȘ­äŒš]DIVERSE TRAJECTORY FORECASTING WITH DETERMINANTAL POINT PROCESSES
[DLèŒȘèȘ­äŒš]DIVERSE TRAJECTORY FORECASTING WITH DETERMINANTAL POINT PROCESSES[DLèŒȘèȘ­äŒš]DIVERSE TRAJECTORY FORECASTING WITH DETERMINANTAL POINT PROCESSES
[DLèŒȘèȘ­äŒš]DIVERSE TRAJECTORY FORECASTING WITH DETERMINANTAL POINT PROCESSES
 
Domain Transfer and Adaptation Survey
Domain Transfer and Adaptation SurveyDomain Transfer and Adaptation Survey
Domain Transfer and Adaptation Survey
 
jamoviă«ă‚ˆă‚‹ăƒ‡ăƒŒă‚żćˆ†æž(1)ïŒšăƒ‡ăƒŒă‚żćˆ†æžăźć‰æ
jamoviă«ă‚ˆă‚‹ăƒ‡ăƒŒă‚żćˆ†æž(1)ïŒšăƒ‡ăƒŒă‚żćˆ†æžăźć‰æjamoviă«ă‚ˆă‚‹ăƒ‡ăƒŒă‚żćˆ†æž(1)ïŒšăƒ‡ăƒŒă‚żćˆ†æžăźć‰æ
jamoviă«ă‚ˆă‚‹ăƒ‡ăƒŒă‚żćˆ†æž(1)ïŒšăƒ‡ăƒŒă‚żćˆ†æžăźć‰æ
 
Causal Inference in Marketing
Causal Inference in MarketingCausal Inference in Marketing
Causal Inference in Marketing
 
Understanding random forests
Understanding random forestsUnderstanding random forests
Understanding random forests
 
Recommender Systems In Industry
Recommender Systems In IndustryRecommender Systems In Industry
Recommender Systems In Industry
 
Counterfactual Learning for Recommendation
Counterfactual Learning for RecommendationCounterfactual Learning for Recommendation
Counterfactual Learning for Recommendation
 
[Paper Reading] Attention is All You Need
[Paper Reading] Attention is All You Need[Paper Reading] Attention is All You Need
[Paper Reading] Attention is All You Need
 
K-Means Clustering Algorithm - Cluster Analysis | Machine Learning Algorithm ...
K-Means Clustering Algorithm - Cluster Analysis | Machine Learning Algorithm ...K-Means Clustering Algorithm - Cluster Analysis | Machine Learning Algorithm ...
K-Means Clustering Algorithm - Cluster Analysis | Machine Learning Algorithm ...
 
Interpretability beyond feature attribution quantitative testing with concept...
Interpretability beyond feature attribution quantitative testing with concept...Interpretability beyond feature attribution quantitative testing with concept...
Interpretability beyond feature attribution quantitative testing with concept...
 
Research of adversarial example on a deep neural network
Research of adversarial example on a deep neural networkResearch of adversarial example on a deep neural network
Research of adversarial example on a deep neural network
 
Learned Embeddings for Search and Discovery at Instacart
Learned Embeddings for  Search and Discovery at InstacartLearned Embeddings for  Search and Discovery at Instacart
Learned Embeddings for Search and Discovery at Instacart
 
Architecture Design for Deep Neural Networks I
Architecture Design for Deep Neural Networks IArchitecture Design for Deep Neural Networks I
Architecture Design for Deep Neural Networks I
 
ă‚ȘăƒŒă‚Żă‚·ăƒ§ăƒłăźç†è«–ăšćźŸè·”2017
ă‚ȘăƒŒă‚Żă‚·ăƒ§ăƒłăźç†è«–ăšćźŸè·”2017ă‚ȘăƒŒă‚Żă‚·ăƒ§ăƒłăźç†è«–ăšćźŸè·”2017
ă‚ȘăƒŒă‚Żă‚·ăƒ§ăƒłăźç†è«–ăšćźŸè·”2017
 
Attention Is All You Need
Attention Is All You NeedAttention Is All You Need
Attention Is All You Need
 
Machine Learning and Inductive Inference
Machine Learning and Inductive InferenceMachine Learning and Inductive Inference
Machine Learning and Inductive Inference
 
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
Rishabh Mehrotra - Recommendations in a Marketplace: Personalizing Explainabl...
 

Ähnlich wie [ICCV 21] Influence-Balanced Loss for Imbalanced Visual Classification

What Makes Training Multi-modal Classification Networks Hard? ppt
What Makes Training Multi-modal Classification Networks Hard? pptWhat Makes Training Multi-modal Classification Networks Hard? ppt
What Makes Training Multi-modal Classification Networks Hard? ppt
taeseon ryu
 
The Impact of Class Rebalancing Techniques on the Performance and Interpretat...
The Impact of Class Rebalancing Techniques on the Performance and Interpretat...The Impact of Class Rebalancing Techniques on the Performance and Interpretat...
The Impact of Class Rebalancing Techniques on the Performance and Interpretat...
Chakkrit (Kla) Tantithamthavorn
 
PROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATION
PROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATIONPROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATION
PROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATION
Willy Marroquin (WillyDevNET)
 
[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification
[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification
[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification
Seulki Park
 
November, 2006 CCKM'06 1
November, 2006 CCKM'06 1 November, 2006 CCKM'06 1
November, 2006 CCKM'06 1
butest
 

Ähnlich wie [ICCV 21] Influence-Balanced Loss for Imbalanced Visual Classification (20)

What Makes Training Multi-modal Classification Networks Hard? ppt
What Makes Training Multi-modal Classification Networks Hard? pptWhat Makes Training Multi-modal Classification Networks Hard? ppt
What Makes Training Multi-modal Classification Networks Hard? ppt
 
Flavours of Physics Challenge: Transfer Learning approach
Flavours of Physics Challenge: Transfer Learning approachFlavours of Physics Challenge: Transfer Learning approach
Flavours of Physics Challenge: Transfer Learning approach
 
Representational Continuity for Unsupervised Continual Learning
Representational Continuity for Unsupervised Continual LearningRepresentational Continuity for Unsupervised Continual Learning
Representational Continuity for Unsupervised Continual Learning
 
The Impact of Class Rebalancing Techniques on the Performance and Interpretat...
The Impact of Class Rebalancing Techniques on the Performance and Interpretat...The Impact of Class Rebalancing Techniques on the Performance and Interpretat...
The Impact of Class Rebalancing Techniques on the Performance and Interpretat...
 
Evoknow17 Large Scale Problems in Practice
Evoknow17 Large Scale Problems in PracticeEvoknow17 Large Scale Problems in Practice
Evoknow17 Large Scale Problems in Practice
 
Boosting auxiliary task guidance: a probabilistic approach
Boosting auxiliary task guidance: a probabilistic approachBoosting auxiliary task guidance: a probabilistic approach
Boosting auxiliary task guidance: a probabilistic approach
 
PROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATION
PROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATIONPROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATION
PROGRESSIVE GROWING OF GAN S FOR I MPROVED QUALITY , STABILITY , AND VARIATION
 
IRJET- Evaluation of Classification Algorithms with Solutions to Class Imbala...
IRJET- Evaluation of Classification Algorithms with Solutions to Class Imbala...IRJET- Evaluation of Classification Algorithms with Solutions to Class Imbala...
IRJET- Evaluation of Classification Algorithms with Solutions to Class Imbala...
 
[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification
[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification
[CVPR 22] Context-rich Minority Oversampling for Long-tailed Classification
 
November, 2006 CCKM'06 1
November, 2006 CCKM'06 1 November, 2006 CCKM'06 1
November, 2006 CCKM'06 1
 
IRJET- Effectiveness of Constructivist Instructional Approach on Achievem...
IRJET-  	  Effectiveness of Constructivist Instructional Approach on Achievem...IRJET-  	  Effectiveness of Constructivist Instructional Approach on Achievem...
IRJET- Effectiveness of Constructivist Instructional Approach on Achievem...
 
An Empirical Study of Training Self-Supervised Vision Transformers.pptx
An Empirical Study of Training Self-Supervised Vision Transformers.pptxAn Empirical Study of Training Self-Supervised Vision Transformers.pptx
An Empirical Study of Training Self-Supervised Vision Transformers.pptx
 
When deep learners change their mind learning dynamics for active learning
When deep learners change their mind  learning dynamics for active learningWhen deep learners change their mind  learning dynamics for active learning
When deep learners change their mind learning dynamics for active learning
 
V.KARTHIKEYAN PUBLISHED ARTICLE
V.KARTHIKEYAN PUBLISHED ARTICLEV.KARTHIKEYAN PUBLISHED ARTICLE
V.KARTHIKEYAN PUBLISHED ARTICLE
 
Learning similarity functions from qualitative feedback
Learning similarity functions from qualitative feedbackLearning similarity functions from qualitative feedback
Learning similarity functions from qualitative feedback
 
result analysis for deep leakage from gradients
result analysis for deep leakage from gradientsresult analysis for deep leakage from gradients
result analysis for deep leakage from gradients
 
Ijetr021251
Ijetr021251Ijetr021251
Ijetr021251
 
IMPROVING SUPERVISED CLASSIFICATION OF DAILY ACTIVITIES LIVING USING NEW COST...
IMPROVING SUPERVISED CLASSIFICATION OF DAILY ACTIVITIES LIVING USING NEW COST...IMPROVING SUPERVISED CLASSIFICATION OF DAILY ACTIVITIES LIVING USING NEW COST...
IMPROVING SUPERVISED CLASSIFICATION OF DAILY ACTIVITIES LIVING USING NEW COST...
 
[Explained] "Partial Success in Closing the Gap between Human and Machine Vis...
[Explained] "Partial Success in Closing the Gap between Human and Machine Vis...[Explained] "Partial Success in Closing the Gap between Human and Machine Vis...
[Explained] "Partial Success in Closing the Gap between Human and Machine Vis...
 
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
IRJET - Factors Affecting Deployment of Deep Learning based Face Recognition ...
 

KĂŒrzlich hochgeladen

+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
?#DUbAI#??##{{(☎+971_581248768%)**%*]'#abortion pills for sale in dubai@
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Safe Software
 

KĂŒrzlich hochgeladen (20)

Real Time Object Detection Using Open CV
Real Time Object Detection Using Open CVReal Time Object Detection Using Open CV
Real Time Object Detection Using Open CV
 
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
 
A Beginners Guide to Building a RAG App Using Open Source Milvus
A Beginners Guide to Building a RAG App Using Open Source MilvusA Beginners Guide to Building a RAG App Using Open Source Milvus
A Beginners Guide to Building a RAG App Using Open Source Milvus
 
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
 
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot TakeoffStrategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
 
Ransomware_Q4_2023. The report. [EN].pdf
Ransomware_Q4_2023. The report. [EN].pdfRansomware_Q4_2023. The report. [EN].pdf
Ransomware_Q4_2023. The report. [EN].pdf
 
Apidays Singapore 2024 - Modernizing Securities Finance by Madhu Subbu
Apidays Singapore 2024 - Modernizing Securities Finance by Madhu SubbuApidays Singapore 2024 - Modernizing Securities Finance by Madhu Subbu
Apidays Singapore 2024 - Modernizing Securities Finance by Madhu Subbu
 
A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
 
Manulife - Insurer Transformation Award 2024
Manulife - Insurer Transformation Award 2024Manulife - Insurer Transformation Award 2024
Manulife - Insurer Transformation Award 2024
 
Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptx
 
Exploring the Future Potential of AI-Enabled Smartphone Processors
Exploring the Future Potential of AI-Enabled Smartphone ProcessorsExploring the Future Potential of AI-Enabled Smartphone Processors
Exploring the Future Potential of AI-Enabled Smartphone Processors
 
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
 
DBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor PresentationDBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor Presentation
 
MS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectorsMS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectors
 
2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...
 
ICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesICT role in 21st century education and its challenges
ICT role in 21st century education and its challenges
 
Navi Mumbai Call Girls đŸ„° 8617370543 Service Offer VIP Hot Model
Navi Mumbai Call Girls đŸ„° 8617370543 Service Offer VIP Hot ModelNavi Mumbai Call Girls đŸ„° 8617370543 Service Offer VIP Hot Model
Navi Mumbai Call Girls đŸ„° 8617370543 Service Offer VIP Hot Model
 
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemkeProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
 

[ICCV 21] Influence-Balanced Loss for Imbalanced Visual Classification

  • 1. MARS: Motion-Augmented RGB Stream for Action Recognition (CVPR 2019) MARS: Motion-Augmented RGB Stream for Action Recognition (CVPR 2019) Influence-Balanced Loss for Imbalanced Visual Classification ASRI, Dept. of Electrical and Computer Engineering Seoul National University Seulki Park Jongin Lim Younghan Jeon Jin Young Choi
  • 2. Imbalanced Visual Classification 2 Introduction Proposed Method Experiment Conclusion Many real-world data often exhibit long-tailed distribution. ✓ The model trained on such imbalanced data tends to overfit the majority classes. ✓ That is, the model performs poorly on minority classes. Problem Definition: ● Input: Long-tailed (imbalanced) training data & uniform-distributed (balanced) test data. ● Goal: To make a robust model that can generalize well on balanced test data. Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021) Faces (Zhang et al., 2017) Places (Wang et al., 2017) Species (Van Horn et al., 2018) Actions (Zhang et al., 2019) * Images by authors.
  • 3. Previous Methods Introduction Proposed Method Experiment Conclusion 1. Data-level approach ◩ Directly balance the training data distributions by re-sampling or generating synthetic samples (Chawla’02, Mullick’ 19, Hulse’ 07). ◩ Under-sampling majority classes can lose some valuable information. ◩ Over-sampling or data generation is susceptible to overfitting to certain repetitive samples. 2. Meta-learning approach ◩ Recent meta-learning methods have shown promising results (Shu’ 19, Liu’20, Ren’20). ◩ However, these methods are difficult to implement in practice. â–Ș Additional unbiased data are required (Shu’19), Meta-sampler is computationally expensive (Ren’20). 3. Re-weighting approach ◩ Assign different weights to each sample according to its importance. ◩ However, they have focused on only global class-level distribution and assign the same weight to all samples belonging to the same class (CB Loss [Cui’19], LDAM [Cao’19]). → “Not all samples in a dataset play an equal role in determining the model parameters. (Cook, 1982)” 3 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
  • 4. Motivation Introduction Proposed Method Experiment Conclusion Q. How can we appropriately re-weight each training samples while preventing the model from overfitting to majority classes? - Focal Loss [Lin’17] assigns more weights to hard examples, but most hard examples belong to majority classes as training progresses. A. Let’s re-weight samples by their direct influences on making the overfitted model! → To measure the influence of a sample, we focus on Influence function (Cook, 1982). 4 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
  • 5. Motivation Introduction Proposed Method Experiment Conclusion Key observations: ✓ Influence of the majority class is much greater than that of the minor class! Key idea: ✓ Down-weight the samples that have large influence on the overfitted decision boundary to make a smoother decision boundary! 5 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021) (a) Comparison of Influences between balanced and imbalanced dataset. (b) key concept of our approach.
  • 6. Proposed Method Introduction Proposed Method Experiment Conclusion 0. Recap: Influence Functions on DNNs (Koh and Liang, ICML, 2017) Let 𝑓(đ‘„, 𝜔) a model, 𝐿(𝑩, 𝑓(đ‘„, 𝜔)) a loss for a training point (đ‘„, 𝑩). By definition, Influence of (đ‘„, 𝑩) on parameters of model (𝜔) is given by: đ›Ș đ‘„; 𝜔 = âˆ’đ»âˆ’1 ∇𝝎𝐿 𝑩, 𝑓 đ‘„, 𝜔 , where đ» ≝ 1 𝑛 Ïƒđ‘–=1 𝑛 ∇𝜔 2 𝐿 𝑩𝑖, 𝑓 đ‘„đ‘–, 𝜔 1. Influence-balanced (IB) weighting factor From đ›Ș đ‘„; 𝜔 , we design the IB weighting factor as follows: IB đ‘„; 𝜔 = ∇𝝎𝐿 𝑩, 𝑓 đ‘„, 𝜔 1 6 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021) (1) (2)
  • 7. Proposed Method Introduction Proposed Method Experiment Conclusion 2. Influence-balanced (IB) Loss When using the softmax cross-entropy loss, IB weighting factor can be further simplified: IB đ‘„; 𝜔 = ∇𝝎𝐿 𝑩, 𝑓 đ‘„, 𝜔 1 = 𝑓 đ‘„, 𝜔 − 𝑩 1 ⋅ ℎ 1 where ℎ is a hidden feature vector. Finally, the influence-balanced loss is given by đżđŒđ”(𝑩, 𝑓 đ‘„, 𝜔 ) = 𝐿(𝑩, 𝑓 đ‘„, 𝜔 ) 𝑓 đ‘„, 𝜔 − 𝑩 1 ⋅ ℎ 1 → The proposed influence-balanced term constrains the decision boundary to not overfit to influential majority samples. 7 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021) (3) (4)
  • 8. Proposed Method Introduction Proposed Method Experiment Conclusion 3. Influence-balanced (IB) Class-wise Re-weighting Finally, we add a class-wise re-weighting term 𝜆𝑘 to the IB-loss in 2. as: đżđŒđ”(𝜔) = 1 𝑚 ෍ đ‘„,𝑩 âˆˆđ·đ‘š 𝜆𝑘 𝐿(𝑩, 𝑓 đ‘„, 𝜔 ) 𝑓 đ‘„, 𝜔 − 𝑩 1 ⋅ ℎ 1 where 𝜆𝑘 = đ›Œ 𝑛𝑘 −1 Ïƒđ‘˜â€Č đŸ 𝑛𝑘â€Č −1. (𝑛𝑘: the number of samples in the 𝑘-th class) → The class-wise re-weighting can further control the influences depending on the classes. 8 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021) (5)
  • 9. Proposed Method Introduction Proposed Method Experiment Conclusion 4. Influence-balanced Training Scheme The influence-balanced training process comprises two phases: 1) normal training and 2) fine-tuning for balance. 9 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
  • 10. Experimental Results Introduction Proposed Method Experiment Conclusion 1. Datasets ◩Synthetic data: â–ȘCIFAR10/100 (10/100 classes), Tiny ImageNet (200 classes) â–ȘLong-tailed imbalance: the number of samples of 𝑘-th class is set to 𝑛𝑘𝜇𝑘, (𝜇 ∈ 0,1 ). â–ȘStep-imbalance: the classes are divided into two groups (majority, minority). ◩ Real-world data: â–ȘiNaturalist 2018: 437,513 images from 8,142 classes (imbalance factor: 500) ※ imbalance factor: the ratio between the most frequent class and the least frequent class. 2. Baselines We compare our method with the following cost-sensitive loss methods: ◩CE: uses standard cross-entropy loss. ◩Focal Loss [Lin et al., ICCV17]: down-weights well-classified samples and up-weights hard samples. ◩CB Loss [Cui et al., CVPR19]: re-weights the loss inversely proportional to the effective number of samples. ◩LDAM [Cao et al., NeurIPS19]: regularizes the minority classes to have larger margins. 10 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
  • 11. Comparison with the state-of-the-arts Introduction Proposed Method Experiment Conclusion Our method achieves the state-of-the-art results on benchmark datasets with various imbalance factors. 11 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021) Classification Accuracy (%) of ResNet-32 on imbalanced CIFAR-10 and CIFAR-100 Classification Accuracy (%) of ResNet-18 on imbalanced Tiny ImageNet. Classification Accuracy (%) of ResNet-50 on iNaturalist2018.
  • 12. Class-wise classification accuracy Introduction Proposed Method Experiment Conclusion Our proposed method shows that the significant performance improvement has resulted from the minority classes, not from the majority classes! 12 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
  • 13. Conclusion Introduction Proposed Method Experiment Conclusion ✓Discovered that the existing loss-based loss methods can lead a decision boundary of DNNs to eventually overfit to the majority classes. ✓Designed a novel influence-balanced loss function to re-weight samples more effectively in a way to alleviate overfitting of decision boundary. ✓Experimentally demonstrated that IB-loss can improve the generalization performance on imbalanced data. ✓Our method is easy to be implemented and integrated into existing methods. 13 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021)
  • 14. Conclusion Introduction Proposed Method Experiment Conclusion ✓Discovered that the existing loss-based loss methods can lead a decision boundary of DNNs to eventually overfit to the majority classes. ✓Designed a novel influence-balanced loss function to re-weight samples more effectively in a way to alleviate overfitting of decision boundary. ✓Experimentally demonstrated that IB-loss can improve the generalization performance on imbalanced data. ✓Our method is easy to be implemented and integrated into existing methods. 14 Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021) Contact: seulki.park@snu.ac.kr Code: https://github.com/pseulki/IB-Loss