Neural trees

Patent metadata

Field	Value
Publication number	US-12175347-B2
Application number	US-202318306888-A
Country	US
Kind code	B2
Filing date	Apr 25, 2023
Priority date	Jun 29, 2018
Publication date	Dec 24, 2024
Grant date	Dec 24, 2024

Abstract

A predictor has a memory which stores at least one example for which an associated outcome is not known. The memory stores at least one decision tree comprising a plurality of nodes connected by edges, the nodes comprising a root node, internal nodes and leaf nodes. Individual ones of the nodes and individual ones of the edges each have an assigned module, comprising parameterized, differentiable operations, such that for each of the internal nodes the module computes a binary outcome for selecting a child node of the internal node. The predictor has a processor configured to compute the prediction by processing the example using a plurality of the differentiable operations selected according to a path through the tree from the root node to a leaf node.

First claim

The invention claimed is:

1. A system comprising:
a memory storing instructions;
a processor configured to execute the instructions stored in the memory to:
create nodes connected by edges at least by computing parameterized, differential operations;
construct, in parallel:
a first model via a growing process dependent on training data; and
a second model by increasing a depth of an incoming edge of at least one of the nodes;

remove at least one of the nodes based on comparing values of the differentiable operations with validation data, resulting in a pruned decision tree; and
apply the pruned decision tree to an input example resulting in a predicted output.

2. The system of claim 1 , wherein the input example is an audio signal, and wherein applying the pruned decision tree comprises predicting a label of an object in the audio signal.

3. The system of claim 1 , wherein the input example is a video file, and wherein applying the pruned decision tree comprises predicting a label of an object in the video file.

4. The system of claim 1 , wherein the input example is a text segment, and wherein applying the pruned decision tree comprises predicting a label of an object in the text segment.

5. The system of claim 1 , wherein the input example is a phoneme from a speech recognition pre-processing system, and wherein applying the pruned decision tree comprises predicting a label of an object in the speech recognition pre-processing system.

6. The system of claim 1 , wherein the input example is a skeletal data, and wherein applying the pruned decision tree comprises predicting a label of an object in the skeletal data.

7. The system of claim 6 , wherein the skeletal data includes estimated skeletal positions of a human or an animal in an image.

8. The system of claim 1 , wherein the input example is sensor data, and wherein applying the pruned decision tree comprises predicting a label of an object in the sensor data.

9. The system of claim 1 , wherein the input example is data derived from sensor data, and wherein applying the pruned decision tree comprises predicting a label of an object in the data.

10. The system of claim 1 , wherein the input example is an image feature map derived from an image, and wherein applying the pruned decision tree comprises predicting a label of an object in the image feature map.

11. The system of claim 1 , wherein the processor is configured to execute the instructions stored in the memory to:
remove, by a transformer on at least one of the edges, noise from the input example.

12. A computer-implemented method comprising:
creating nodes connected by edges at least by computing parameterized, differential operations;
constructing, in parallel:
a first model via a growing process dependent on training data; and
a second model by increasing a depth of an incoming edge of at least one of the nodes;

removing at least one of the nodes based on comparing values of the differentiable operations with validation data, resulting in a pruned decision tree; and
applying the pruned decision tree to an input example resulting in a predicted output.

13. The computer-implemented method of claim 12 , wherein the input example is an audio signal, and wherein applying the pruned decision tree comprises predicting a label of an object in the audio signal.

14. The computer-implemented method of claim 12 , wherein the input example is a video file, and wherein applying the pruned decision tree comprises predicting a label of an object in the video file.

15. The computer-implemented method of claim 12 , wherein the input example is a text segment, and wherein applying the pruned decision tree comprises predicting a label of an object in the text segment.

16. The computer-implemented method of claim 12 , wherein the input example is a phoneme from a speech recognition pre-processing system, and wherein applying the pruned decision tree comprises predicting a label of an object in the speech recognition pre-processing system.

17. The computer-implemented method of claim 12 , wherein the input example is a skeletal data, wherein applying the pruned decision tree comprising predicting a label of an object in the skeletal data, and wherein the skeletal data is estimated skeletal positions of a human or an animal in an image.

18. The computer-implemented method of claim 12 , wherein the input example is sensor data, and wherein applying the pruned decision tree comprises predicting a label of an object in the sensor data.

19. The computer-implemented method of claim 12 , wherein the input example is data derived from sensor data, and wherein applying the pruned decision tree comprises predicting a label of an object in the data.

20. A server comprising:
a memory storing instructions;
a processor configured to execute the instructions stored in the memory to:
create nodes connected by edges at least by computing parameterized, differential operations;
construct, in parallel:
a first model via a growing process dependent on training data; and
a second model by increasing a depth of an incoming edge of at least one of the nodes;

remove at least one of the nodes based on comparing values of the differentiable operations with validation data, resulting in a pruned decision tree; and
apply the pruned decision tree to an input example resulting in a predicted output.

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Nori Aditya Vithal
• Criminisi Antonio
• TANNO Ryutaro

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0985

  Hyperparameter optimisation; Meta-learning; Learning-to-learn · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title