Electronic apparatus and method for re-learning trained model

1. A method for re-learning a trained model, comprising:

receiving an input trained model consisting of a plurality of neurons and a data set including a new task;

identifying a neuron associated with the new task among the plurality of neurons of the input trained model and selectively re-learning a parameter associated with the new task for the identified neuron, wherein the selectively re-learning only re-learns the identified neuron associated with the new task; and

reconstructing the input trained model by adding one or more neurons to a selectively re-learned trained model for dynamically expanding a size of the selectively re-learned trained model on which the selective re-learning is performed to a second size greater than a first size of the input trained model if a loss of the selectively re-learned trained model exceeds a preset loss value,

wherein in the reconstructing of the input trained model, when the loss exceeds the preset loss value, the one or more neurons comprising a fixed number of neurons for each layer is added to the selectively re-learned trained model and group sparsity is used to eliminate unnecessary neurons from the added neurons, thereby reconstructing the input trained model,

wherein in the reconstructing of the input trained model, an unnecessary neuron is identified from the added neurons using an objective function having a loss function for the input trained model, a regularization term for sparsity, and a group regularization term for group sparsity,

wherein in the objective function having loss function (IL) for the input trained model, the regularization term for sparsity is based on an L1 norm and the group regularization term for the group sparsity is based on an L2 norm as follows:

min

W

l

𝕃

⁡

(

W

l

;

W

l

t

-

1

,

D

t

)

+

μ

⁢



W

l



1

+

γ

⁢

∑

g



W

l

,

g



2

,

wherein W are the neural network weights, W has L layers indexed by variable 1 , t is the current task for which W is being updated, Dt represents data for the current task t, μ is a first hyperparameter, γ is a second hyperparameter, and g represents a group defined as the inflow weights for each neuron.

2. The method as claimed in claim 1 , wherein in the selective re-learning, a new parameter matrix is calculated using the data set for a network parameter consisting of only the identified neuron, and the calculated new parameter matrix is reflected to the identified neuron of the trained model to perform the selective re-learning.

3. The method as claimed in claim 1 , wherein in the reconstructing of the input trained model, if a change in the identified neuron has a preset value, the identified neuron is duplicated to expand the input trained model, and the identified neuron has an existing value to reconstruct the input trained model.

4. The method as claimed in claim 1 , further comprising:

limiting the size of the input trained model based on a cumulative knowledge accounting of the new task.

5. A method of learning a new concept using already-learned knowledge, the method comprising:

training a model to provide a first trained model, the first trained model then being the already-learned knowledge;

classifying first data using the first trained model to provide a first classification;

re-learning the first trained model using the method of claim 1 to provide a second trained model; and

classifying second data using the second trained model to provide a second classification, wherein the second classification corresponds to the new concept.