Back to results
US 9,805,305 B2Granted
Boosted deep convolutional neural networks (CNNs)
AssigneeYahoo Holdings, Inc.
InventorsMohammad Saberian, Jia Li, Jian Yang
Life of this patent
Estimating the term…
Abstract
Briefly, embodiments of methods and/or systems of training multiclass convolutional neural networks (CNNs) are disclosed. For one embodiment, as an example, an auxiliary CNN may be utilized to form an ensemble with the collection as a linear combination. The linear combination may be based, at least in part, on boost prediction error encountered during the training process.
Claims20
A method comprising:
- training a collection of multiclass convolutional neural networks (CNNs) via a boosting process comprising at least one boost iteration to utilize an auxiliary CNN, the auxiliary CNN to form an ensemble with the collection as a linear combination, the linear combination based at least in part on boost prediction error encountered during the training with a subset of the collection.
An apparatus, comprising:
- one or more processors to:
- train a collection of multiclass convolutional neural networks (CNNs) via a boosting process to comprise at least one boost iteration to utilize an auxiliary CNN, the auxiliary CNN to form an ensemble with the collection as a linear combination, the linear combination based at least in part on boost prediction error to be encountered during the training with a subset of the collection.
An apparatus comprising:
- means for generating boost prediction error;
- and means for training a collection of multiclass convolutional neural networks (CNNs) via an iterative boosting process to utilize an auxiliary CNN, the auxiliary CNN to form an ensemble with the collection as a linear combination, the linear combination based at least in part on means for generating boost prediction error.
Dependencies are inferred from the claim text.