Dynamic content encoding
A method for encoding text includes grouping text as a sequence of bytes, the text comprising a string of characters, each byte corresponding to a character in the text. For each byte of the sequence of bytes: (a) each bit is processed from most significant bit to least significant bit to generate a context; and (b) a subsequent bit is predicted, using a prediction model, based on the context generated based on previously processed bits, prediction of the prediction model being a combination of predictions of a plurality of sub-models. An encoded bitstream is output based on the predicted bits. The encoded bitstream includes encoded data corresponding to the text.
1. A method, comprising:
grouping text as a sequence of bytes, the text comprising a string of characters, each byte corresponding to a character in the text;
for each byte of the sequence of bytes:
processing each bit from most significant bit to least significant bit to generate a context; and
predicting a subsequent bit, using a prediction model, based on the context generated based on previously processed bits, prediction of the prediction model being a combination of predictions of a plurality of sub-models; and
outputting an encoded bitstream based on the predicted bits, the encoded bitstream comprising encoded data corresponding to the text.
2. The method of claim 1 , wherein predicting the subsequent bit comprises generating a probability, using each of the plurality of sub-models, that the subsequent bit is 1 using a corresponding pre-determined number of previously processed bits as context, the probability being determined by a hash table lookup for the corresponding sub-model.
3. The method of claim 1 , wherein at least one of the plurality of sub-models predicts a subsequent character based on a previously processed string of characters in the text.
4. The method of claim 1 , wherein each of the plurality of sub-models is static model and is pre-trained using a known data set.
5. The method of claim 1 , wherein the combination of predictions of the plurality of sub-models comprises combining different order predictions using a 3D function, each of the different order predictions obtained by combining a corresponding subset of sub-models among the plurality of sub-models.
6. The method of claim 5 , wherein a low order prediction is based on a sub-model that uses one previously processed bit as a context for predicting the subsequent bit.
7. The method of claim 5 , wherein combining the predictions of a first subset is performed using a medium order 3D function, and combining the predictions of a second subset is performed using a high order 3D function.
8. The method of claim 1 , wherein the outputted encoded bitstream comprises a first encoded portion indicating a size of the data being encoded, a single bit indicating whether the data is compressed, and a third encoded portion encoding the data corresponding to the text.
9. A non-transitory computer-readable medium including instructions, which when executed by at least one computing device, cause the at least one computing device to perform operations including:
grouping text as a sequence of bytes, the text comprising a string of characters, each byte corresponding to a character in the text;
for each byte of the sequence of bytes:
processing each bit from most significant bit to least significant bit to generate a context; and
predicting a subsequent bit, using a prediction model, based on the context generated based on previously processed bits, prediction of the prediction model being a combination of predictions of a plurality of sub-models; and
outputting an encoded bitstream based on the predicted bits, the encoded bitstream comprising encoded data corresponding to the text.
10. The non-transitory computer-readable medium of claim 9 , wherein predicting the subsequent bit comprises generating a probability, using each of the plurality of sub-models, that the subsequent bit is 1 using a corresponding pre-determined number of previously processed bits as context, the probability being determined by a hash table lookup for the corresponding sub-model.
11. The non-transitory computer-readable medium of claim 9 , wherein at least one of the plurality of sub-models predicts a subsequent character based on a previously processed string of characters in the text.
12. The non-transitory computer-readable medium of claim 9 , wherein each of the plurality of sub-models is static model and is pre-trained using a known data set.
13. The non-transitory computer-readable medium of claim 9 , wherein the combination of predictions of the plurality of sub-models comprises combining different order predictions using a 3D function, each of the different order predictions obtained by combining a corresponding subset of sub-models among the plurality of sub-models.
14. The non-transitory computer-readable medium of claim 13 , wherein a low order prediction is based on a sub-model that uses one previously processed bit as a context for predicting the subsequent bit.
15. The non-transitory computer-readable medium of claim 13 , wherein combining the predictions of a first subset is performed using a medium order 3D function, and combining the predictions of a second subset is performed using a high order 3D function.
16. The non-transitory computer-readable medium of claim 9 , wherein the outputted encoded bitstream comprises a first encoded portion indicating a size of the data being encoded, a single bit indicating whether the data is compressed, and a third encoded portion encoding the data corresponding to the text.
17. A device including an input, an output, a memory and at least one processor communicatively coupled to the memory, the at least one processor configured to:
group text as a sequence of bytes, the text comprising a string of characters, each byte corresponding to a character in the text;
for each byte of the sequence of bytes:
process each bit from most significant bit to least significant bit to generate a context; and
predict a subsequent bit, using a prediction model, based on the context generated based on previously processed bits, prediction of the prediction model being a combination of predictions of a plurality of sub-models; and
output an encoded bitstream based on the predicted bits, the encoded bitstream comprising encoded data corresponding to the text.
18. The device of claim 17 , wherein predicting the subsequent bit comprises generating a probability, using each of the plurality of sub-models, that the subsequent bit is 1 using a corresponding pre-determined number of previously processed bits as context, the probability being determined by a hash table lookup for the corresponding sub-model.
19. The device of claim 17 , wherein the combination of predictions of the plurality of sub-models comprises combining different order predictions using a 3D function, each of the different order predictions obtained by combining a corresponding subset of sub-models among the plurality of sub-models.
20. The device of claim 17 , wherein the outputted encoded bitstream comprises a first encoded portion indicating a size of the data being encoded, a single bit indicating whether the data is compressed, and a third encoded portion encoding the data corresponding to the text.