SYSTEM TO CLUSTER SLEEPERS WITH FEATURES FOR EMBEDDING OBJECTIVE DATA AND TEXT DATA
A plurality of beds each have, one or more sensors configured to: sense physical phenomena of a user of the bed; and send, to a computer system, sensor readings, wherein the sensor readings include objective data representing measurements of the physical phenomena. A computer system is configured to: receive the sensor readings; receive a plurality of text objects; embed the sensor readings into first vectors of a fixed size; embedding each of the text objects into second vectors of the fixed size; generating, for each user, an aggregated matrix that represents objective data and plurality of text objects; generating a clustering of the users based on the objective data and the plurality of text objects such that users with similar data are placed into the same cluster; and generating a visual representation describing the clustering of the user users.
1 . A system comprising:
a plurality of beds;
for each bed, one or more sensors configured to:
sense physical phenomena of a user of the bed; and
send, to a computer system, sensor readings, wherein the sensor readings comprise objective data representing measurements of the physical phenomena; and
a computer system comprising memory and one or more processors, the computer system configured to:
receive the sensor readings;
receive a plurality of text objects entered by the user;
embed, using a quantizing engine, the sensor readings into first vectors of a fixed size, the first vectors representing each of the measurements of the physical phenomena;
embed, using a transformer model, each of the text objects entered by the user into second vectors of the fixed size, the second vectors representing each text object;
generate, for each user, using the first and second vectors, an aggregated matrix that represents objective data and plurality of text objects;
generate, using the aggregated matrices, a clustering of the users based on the objective data and the plurality of text objects such that users with similar data are placed into the same cluster; and
generate, using the clustering, a visual representation describing the clustering of the users.
2 . The system of claim 1 , wherein the objective data comprises data selected from the the group consisting of (i) waveforms, (ii) vectors, (iii) scalars, and (iv) images.
3 . The system of claim 2 , wherein the waveforms comprise electrophysiological signals.
4 . The system of claim 2 , wherein the vectors comprise sleep data selected from the group consisting of (i) duration of sleep, (ii) duration of sleep stages, and (iii) sleep latencies.
5 . The system of claim 2 , wherein the scalars comprise values selected from the group consisting of (i) single categorical values, (ii) ordinal values, and (iii) numeric values that represent age and gender of the user.
6 . The system of claim 2 , wherein embedding, using a quantization engine, a waveform into a first vector of a fixed size comprises:
processing the waveform using a bandpass filter to generate a filtered waveform;
generating a scaled waveform comprising scaling the filtered waveform such that an amplitude of the scaled waveform does not exceed a predetermined threshold; and
mapping, using quantization techniques, the scaled waveform to a vector of the fixed size.
7 . The system of claim 2 , wherein embedding, using a quantization engine, a vector into a first vector of a fixed size comprises:
mapping, using quantization techniques, the vector to a vector of the fixed size.
8 . The system of claim 7 , wherein embedding, using a quantization engine, a vector into a first vector of a fixed size further comprises:
generating a summarization of the vector comprising analyzing components of the vector.
9 . The system of claim 2 , wherein embedding, using a quantization engine, a scalar into a first vector of a fixed size comprises:
mapping, using quantization techniques, the scalar to a vector of the fixed size.
10 . The system of claim 2 , wherein embedding, using a quantization engine, an image into a first vectors of a fixed size comprises:
identifying regions of interest comprising segmenting the image;
generating a summary vector comprising summarizing the regions of interest; and
mapping, using quantization techniques, the summary vector to a vector of the fixed size.
11 . The system of claim 10 , wherein the image comprises a picture of the user in the bed and wherein segmenting the image further comprises identifying regions that include a presence of the user and regions that include an absence of the user.
12 . The system of claim 1 , wherein the one or more sensors include sensors that measure data selected from the group consisting of (i) electrophysiological signals, (ii) sleep architecture data, and (iii) biometrics data.
13 . The system of claim 1 , wherein embedding, using a transformer model, each of the text objects entered by the user into second vectors of the fixed size, the second vectors representing each text object comprises:
mapping the text objects into a vector space that preserves semantic similarity.
14 . The system of claim 13 , wherein mapping the text objects into a vector space that preserves semantic similarity comprises segmenting the text object into a plurality of smaller-sized text objects.
15 . The system of claim 1 , wherein the plurality of text objects entered by the user comprises data selected from the group consisting of (i) subjective feedback about the user's sleep, and (ii) diagnostic information.
16 . The system of claim 1 , wherein the computer system is further configured to:
receive supplemental text objects that describe the objective data; and
embed, using a transformer model, each of the supplemental text objects entered by the user into second vectors of the fixed size, the second vectors representing each supplemental text object.
17 . The system of claim 1 , wherein generating, using the aggregated matrix, the clustering of the users comprises using a K means algorithm.
18 . The system of claim 17 , wherein the using the K means algorithm comprises minimizing a distortion score.
19 . A method comprising, for each bed of a plurality of beds:
receiving, from one or more sensors configured to sense physical phenomena of a user of the bed, sensor readings comprising objective data representing measurements of the physical phenomena;
receiving a plurality of text objects entered by the user;
embedding, using a quantizing engine, the sensor readings into first vectors of a fixed size, the first vectors representing each of the measurements of the physical phenomena;
embedding, using a transformer model, each of the text objects entered by the user into second vectors of the fixed size, the second vectors representing each text object;
generating, for each user, using the first and second vectors, an aggregated matrix that represents objective data and plurality of text objects;
generating, using the aggregated matrices, a clustering of the users based on the objective data and the plurality of text objects such that users with similar data are placed into the same cluster; and
generating, using the clustering, a visual representation describing the clustering of the users. One or more non-transitory machine-readable medium storing instructions that, when executed, are configured to cause one or more processors to perform operations comprising, for each bed of a plurality of beds:
receiving, from one or more sensors configured to sense physical phenomena of a user of the bed, sensor readings comprising objective data representing measurements of the physical phenomena;
receiving a plurality of text objects entered by the user;
embedding, using a quantizing engine, the sensor readings into first vectors of a fixed size, the first vectors representing each of the measurements of the physical phenomena;
embedding, using a transformer model, each of the text objects entered by the user into second vectors of the fixed size, the second vectors representing each text object;
generating, for each user, using the first and second vectors, an aggregated matrix that represents objective data and plurality of text objects;
generating, using the aggregated matrices, a clustering of the users based on the objective data and the plurality of text objects such that users with similar data are placed into the same cluster; and
generating, using the clustering, a visual representation describing the clustering of the users.