Women also Snowboard: Overcoming Bias in Captioning Models

Abstract

Most machine learning methods are known to capture and exploit biases of the training data. While some biases are beneficial for learning, others are harmful. Specifically, image captioning models tend to exaggerate biases present in training data (e.g., if a word is present in 60% of training sentences, it might be predicted in 70% of sentences at test time). This can lead to incorrect captions in domains where unbiased captions are desired, or required, due to over-reliance on the learned prior and image context. In this work we investigate generation of gender-specific caption words (e.g. man, woman) based on the person's appearance or the image context. We introduce a new Equalizer model that ensures equal gender probability when gender evidence is occluded in a scene and confident predictions when gender evidence is present. The resulting model is forced to look at a person rather than use contextual cues to make a gender-specific predictions. The losses that comprise our model, the Appearance Confusion Loss and the Confident Loss, are general, and can be added to any description model in order to mitigate impacts of unwanted bias in a description dataset. Our proposed model has lower error than prior work when describing images with people and mentioning their gender and more closely matches the ground truth ratio of sentences including women to sentences including men. We also show that unlike other approaches, our model is indeed more often looking at people when predicting their gender.

arXiv comment22 pages, 6 figures, Burns and Hendricks contributed equally

References in corpus (8)

• Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
• Equality of Opportunity in Supervised Learning
• Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
• Visualizing Deep Neural Network Decisions: Prediction Difference Analysis
• Deep Transfer Network: Unsupervised Domain Adaptation
• InclusiveFaceNet: Improving Face Attribute Detection with Race and Gender Diversity
• Stereotyping and Bias in the Flickr30K Dataset
• ConvNets and ImageNet Beyond Accuracy: Understanding Mistakes and Uncovering Biases

Cited by in corpus (47)

• RUBi: Reducing Unimodal Biases in Visual Question Answering
• Ground Truth Evaluation of Neural Network Explanations with CLEVR-XAI
• Image Representations Learned With Unsupervised Pre-Training Contain Human-like Biases
• Large-Scale Long-Tailed Recognition in an Open World
• Towards Accountability for Machine Learning Datasets: Practices from Software Engineering and Infrastructure
• Mitigating Gender Bias in Natural Language Processing: Literature Review
• Deep Learning for Face Recognition: Pride or Prejudiced?
• Unbiased Scene Graph Generation from Biased Training
• Unshuffling Data for Improved Generalization
• Interpretations are useful: penalizing explanations to align neural networks with prior knowledge
• Towards Explainable Neural-Symbolic Visual Reasoning
• Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations
• Don't Take the Easy Way Out: Ensemble Based Methods for Avoiding Known Dataset Biases
• Diverse Image Captioning with Context-Object Split Latent Spaces
• iPerceive: Applying Common-Sense Reasoning to Multi-Modal Dense Video Captioning and Video Question Answering
• Reducing Sentiment Bias in Language Models via Counterfactual Evaluation
• Taking a Stance on Fake News: Towards Automatic Disinformation Assessment via Deep Bidirectional Transformer Language Models for Stance Detection
• Directional Bias Amplification
• Learning Not to Learn: Training Deep Neural Networks with Biased Data
• Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation
• FairFaceGAN: Fairness-aware Facial Image-to-Image Translation
• U-Noise: Learnable Noise Masks for Interpretable Image Segmentation
• Overcoming Language Priors in Visual Question Answering with Adversarial Regularization
• Auditing ImageNet: Towards a Model-driven Framework for Annotating Demographic Attributes of Large-Scale Image Datasets
• Are You Looking? Grounding to Multiple Modalities in Vision-and-Language Navigation
• More Grounded Image Captioning by Distilling Image-Text Matching Model
• Estimating semantic structure for the VQA answer space
• Exposing and Correcting the Gender Bias in Image Captioning Datasets and Models
• Adaptive Label Smoothing
• EnD: Entangling and Disentangling deep representations for bias correction
• A non-discriminatory approach to ethical deep learning
• Jigsaw-VAE: Towards Balancing Features in Variational Autoencoders
• Learning Credible Deep Neural Networks with Rationale Regularization
• iReason: Multimodal Commonsense Reasoning using Videos and Natural Language with Interpretability
• Weighted Empirical Risk Minimization: Sample Selection Bias Correction based on Importance Sampling
• The Geometry of Distributed Representations for Better Alignment, Attenuated Bias, and Improved Interpretability
• A Set of Distinct Facial Traits Learned by Machines Is Not Predictive of Appearance Bias in the Wild
• Mitigating Gender Bias in Captioning Systems
• Person Perception Biases Exposed: Revisiting the First Impressions Dataset
• Grounded Video Description
• Benchmarking CNN on 3D Anatomical Brain MRI: Architectures, Data Augmentation and Deep Ensemble Learning
• Hide-and-Tell: Learning to Bridge Photo Streams for Visual Storytelling
• Mitigating Gender Bias Amplification in Distribution by Posterior Regularization
• Unsupervised Learning of Debiased Representations with Pseudo-Attributes
• Perceptual Score: What Data Modalities Does Your Model Perceive?
• Debiasing Convolutional Neural Networks via Meta Orthogonalization
• Diagnostic Visualization for Deep Neural Networks Using Stochastic Gradient Langevin Dynamics