Selecting a subset of training data from a data pool for a power prediction model
A method includes generating a plurality of vector sequences based on input signals of an electric circuit design and encoding the plurality of vector sequences. The method also includes clustering the plurality of encoded vector sequences into a plurality of clusters and selecting a set of encoded vector sequences from the plurality of clusters. The method further includes selecting a first set of vector sequences corresponding to the selected set of encoded vector sequences, selecting a second set of vector sequences from the plurality of vector sequences not in the first set of encoded vector sequences, and training, by a processing device, a machine learning model to predict power consumption using the first and second sets of vector sequences.
1. A method comprising:
generating a plurality of vector sequences based on input signals to an electric circuit design;
encoding the plurality of vector sequences;
clustering the plurality of encoded vector sequences into a plurality of clusters;
selecting a set of encoded vector sequences from the plurality of clusters;
selecting a first set of vector sequences corresponding to the selected set of encoded vector sequences;
selecting a second set of vector sequences from the plurality of vector sequences not in the first set of vector sequences; and
training, by a processing device, a machine learning model to predict power consumption using the first and second sets of vector sequences.
2. The method of claim 1 , further comprising predicting, using the machine learning model, a power consumption of the electric circuit design based on a set of vector sequences in the plurality of vector sequences that are not in the first and second sets of vector sequences.
3. The method of claim 1 , wherein selecting the set of encoded vector sequences comprises selecting, from each cluster of the plurality of clusters, a number of encoded vector sequences nearest to a centroid of that cluster.
4. The method of claim 1 , wherein selecting the second set of vector sequences comprises iteratively selecting, using query-by-committee (QBC), a batch of vector sequences.
5. The method of claim 1 , wherein selecting the second set of vector sequences comprises:
selecting two batches of vector sequences;
clustering input features of the two batches into a second plurality of clusters; and
selecting half of the two batches of vector sequences based on the second plurality of clusters.
6. The method of claim 1 , wherein selecting the second set of vector sequences comprises:
selecting, using query-by-committee, two batches of vector sequences;
clustering embeddings or hidden representations of the two batches into a second plurality of clusters; and
selecting half of the two batches of vector sequences based on the second plurality of clusters.
7. The method of claim 1 , wherein a total number of the vector sequences in the first and second sets of vector sequences is less than a total number of vector sequences in the plurality of vector sequences.
8. The method of claim 1 , further comprising generating a plurality of input vectors based on the input signals to the electric circuit design, wherein each input vector of the plurality of input vectors comprises values of the input signals to the electric circuit design over a cycle period of the electric circuit design.
9. The method of claim 1 , further comprising generating a plurality of input vectors based on the input signals to the electric circuit design, wherein each vector sequence of the plurality of vector sequences comprises a subset of the plurality of input vectors arranged chronologically.
10. The method of claim 1 , further comprising determining a power consumption of the electric circuit design for each vector sequence in the first set of vector sequences to produce a set of power consumption datapoints, wherein training the machine learning model further uses the set of power consumption datapoints.
11. The method of claim 1 , wherein the plurality of vector sequences are encoded using a neural network and an encoder.
12. A system comprising:
a memory; and
a hardware processor communicatively coupled to the memory, the hardware processor configured to:
generate a plurality of vector sequences based on input signals to an electric circuit design;
encode the plurality of vector sequences;
cluster the plurality of encoded vector sequences into a plurality of clusters;
select a set of encoded vector sequences from the plurality of clusters;
select a first set of vector sequences corresponding to the selected set of encoded vector sequences;
select a second set of vector sequences from the plurality of vector sequences not in the first set of vector sequences; and
train a machine learning model to predict power consumption using the first and second sets of vector sequences.
13. The system of claim 12 , wherein the hardware processor is further configured to predict, using the machine learning model, a power consumption of the electric circuit design based on a set of vector sequences in the plurality of vector sequences that are not in the first and second sets of vector sequences.
14. The system of claim 12 , wherein selecting the second set of vector sequences comprises iteratively selecting, using query-by-committee, a batch of vector sequences.
15. The system of claim 12 , wherein selecting the second set of vector sequences comprises:
selecting two batches of vector sequences;
clustering input features of the two batches into a second plurality of clusters; and
selecting half of the two batches of vector sequences based on the second plurality of clusters.
16. The system of claim 12 , wherein selecting the second set of vector sequences comprises:
selecting, using query-by-committee, two batches of vector sequences;
clustering embeddings or hidden representations of the two batches into a second plurality of clusters; and
selecting half of the two batches of vector sequences based on the second plurality of clusters.
17. A non-transitory computer readable medium comprising software instructions that, when executed, cause a processor to perform an operation comprising:
generating a plurality of vector sequences based on input signals to an electric circuit design;
clustering the plurality of vector sequences into a plurality of clusters;
selecting a first set of vector sequences from the plurality of clusters;
selecting a second set of vector sequences from the plurality of vector sequences not in the first set of vector sequences; and
training a machine learning model to predict power consumption using the first and second sets of vector sequences.
18. The non-transitory computer readable medium of claim 17 , the operation further comprising predicting, using the machine learning model, a power consumption of the electric circuit design based on a set of vector sequences in the plurality of vector sequences that are not in the first and second sets of vector sequences.
19. The non-transitory computer readable medium of claim 17 , wherein selecting the second set of vector sequences comprises iteratively selecting, using query-by-committee, a batch of vector sequences.