The Limited Multi-Label Projection Layer
arXiv:1906.08707
Abstract
We propose the Limited Multi-Label (LML) projection layer as a new primitive operation for end-to-end learning systems. The LML layer provides a probabilistic way of modeling multi-label predictions limited to having exactly k labels. We derive efficient forward and backward passes for this layer and show how the layer can be used to optimize the top-k recall for multi-label tasks with incomplete label information. We evaluate LML layers on top-k CIFAR-100 classification and scene graph generation. We show that LML layers add a negligible amount of computational overhead, strictly improve the model's representational capacity, and improve accuracy. We also revisit the truncated top-k entropy method as a competitive baseline for top-k classification.
References in corpus (17)
- From Softmax to Sparsemax: A Sparse Model of Attention and Multi-Label Classification
- Pixels to Graphs by Associative Embedding
- Learning Latent Permutations with Gumbel-Sinkhorn Networks
- Mapping Images to Scene Graphs with Permutation-Invariant Structured Prediction
- Ranking via Sinkhorn Propagation
- Discovering objects and their relations from entangled scene representations
- Smooth Loss Functions for Deep Top-k Classification
- SparseMAP: Differentiable Sparse Structured Inference
- Top Rank Optimization in Linear Time
- On Support Relations and Semantic Scene Graphs
- Top-k Multiclass SVM
- Structured Prediction Energy Networks
- Sparse Support Vector Infinite Push
- Transductive Optimization of Top k Precision
- Structured Prediction with Projection Oracles
- Predict and Constrain: Modeling Cardinality in Deep Structured Prediction
- Learning with Fenchel-Young Losses
Cited by in corpus (5)
- Decision-Focused Learning: Foundations, State of the Art, Benchmark and Future Opportunities
- Gradient Estimation with Stochastic Softmax Tricks
- Backpropagation of Unrolled Solvers with Folded Optimization
- The Differentiable Cross-Entropy Method
- NOVAS: Non-convex Optimization via Adaptive Stochastic Search for End-to-End Learning and Control