Meta-learning for multi-task learning for neural networks
Methods and systems for meta-learning are described for automating learning of child tasks with a single neural network. The order in which tasks are learned by the neural network can affect performance of the network, and the meta-learning approach can use a task-level curriculum for multi-task training. The task-level curriculum can be learned by monitoring a trajectory of loss functions during training. The meta-learning approach can learn to adapt task loss balancing weights in the course of training to get improved performance on multiple tasks on real world datasets. Advantageously, learning to dynamically balance weights among different task losses can lead to superior performance over the use of static weights determined by expensive random searches or heuristics. Embodiments of the meta-learning approach can be used for computer vision tasks or natural language processing tasks, and the trained neural networks can be used by augmented or virtual reality devices.
1. A system for training a neural network to learn a set of tasks, the system comprising:
non-transitory memory configured to
store: executable instructions;
and
a child network for learning a plurality of tasks, wherein the child network is associated with a loss function for the plurality of tasks and a task weight is assigned to each task of the plurality of tasks;
a hardware processor in communication with the non-transitory memory, the hardware processor programed by the executable instructions to:
determine a first child network loss associated with the loss function of the child network;
determine an updated task weight for each task of the plurality of tasks based on the first child network loss;
determine an updated child network based on the updated task weight for each task of the plurality of tasks;
determine a second child network loss associated with the loss function of the updated child network; and
determine a second updated task weight for each task of the plurality of tasks based at least on the second child network loss.
2. The system of claim 1 , wherein the hardware processor is further programmed to learn a relationship between the first and second child network loss.
3. The system of claim 2 , wherein the hardware processor is programmed to determine the second updated task weight based on at least the relationship between the first and second child network loss.
4. The system of claim 1 , wherein the plurality of tasks comprise at least semantic segmentation, depth prediction, and surface normal estimation.
5. The system of claim 1 , wherein the plurality of tasks comprises a regression task, a classification task, or a combination thereof.
6. The system of claim 5 wherein the classification task comprises perception, face recognition, visual search, gesture recognition, semantic segmentation, object detection, room layout estimation, cuboid detection, lighting detection, simultaneous localization and mapping, relocalization, speech processing, speech recognition, natural language processing, or a combination.
7. The system of claim 1 , wherein the loss function comprises a weighted linear combination of losses associated with the plurality of tasks based on task weights corresponding to the plurality of tasks.
8. A method implemented by a system of one or more processors, the method comprising:
determining a first child network loss associated with a loss function of a child network, wherein the child network is configured for learning a plurality of tasks, wherein the child network is associated with the loss function for the plurality of tasks and a task weight is assigned to each task of the plurality of tasks;
determining an updated task weight for each task of the plurality of tasks based on the first child network loss;
determining an updated child network based on the updated task weight for each task of the plurality of tasks;
determining a second child network loss associated with the loss function of the updated child network; and
determining a second updated task weight for each task of the plurality of tasks based at least on the second child network loss.
9. The method of claim 8 , wherein the method further comprises determining a relationship between the first and second child network loss.
10. The method of claim 9 , wherein determining the second updated task weight is based on at least the relationship between the first and second child network loss.
11. The method of claim 8 , wherein the plurality of tasks comprise at least semantic segmentation, depth prediction, and surface normal estimation.
12. The method of claim 8 , wherein the plurality of tasks comprises a regression task, a classification task, or a combination thereof.
13. The method of claim 12 , wherein the classification task comprises perception, face recognition, visual search, gesture recognition, semantic segmentation, object detection, room layout estimation, cuboid detection, lighting detection, simultaneous localization and mapping, relocalization, speech processing, speech recognition, natural language processing, or a combination.
14. The method of claim 8 , wherein the loss function comprises a weighted linear combination of losses associated with the plurality of tasks based on task weights corresponding to the plurality of tasks.
15. Non-transitory computer storage media storing instructions which when executed by a system of one or more processors, cause the system to perform operations comprising:
determining a first child network loss associated with a loss function of a child network, wherein the child network is configured for learning a plurality of tasks, wherein the child network is associated with the loss function for the plurality of tasks and a task weight is assigned to each task of the plurality of tasks;
determining an updated task weight for each task of the plurality of tasks based on the first child network loss;
determining an updated child network based on the updated task weight for each task of the plurality of tasks;
determining a second child network loss associated with the loss function of the updated child network; and
determining a second updated task weight for each task of the plurality of tasks based at least on the second child network loss.
16. The computer storage media of claim 15 , wherein the operations further comprise determining a relationship between the first and second child network loss.
17. The computer storage media of claim 16 , wherein determining the second updated task weight is based on at least the relationship between the first and second child network loss.
18. The computer storage media of claim 15 , wherein the plurality of tasks comprise at least semantic segmentation, depth prediction, and surface normal estimation.
19. The computer storage media of claim 15 , wherein the plurality of tasks comprises a regression task, a classification task, or a combination thereof.
20. The computer storage media of claim 19 , wherein the classification task comprises perception, face recognition, visual search, gesture recognition, semantic segmentation, object detection, room layout estimation, cuboid detection, lighting detection, simultaneous localization and mapping, relocalization, speech processing, speech recognition, natural language processing, or a combination.