# US 2026/0310972 A1 · Improved training of slot-based machine learning models

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0310972 A1
- Document ID: US20260310972A1
- Status: Published application
- Kind: A1
- Application number: 19/101,022
- Filed: 2023-09-21
- Published: 2026-10-08
- Earliest claimed: 2022-09-21
- Applicants: Robert Bosch GmbH
- Inventors: Ning Gao; Gerhard Neumann
- Web page: https://patentgrep.com/patent/US20260310972A1

## CPC

- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/048: Activation functions
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method for training a machine learning model that is configured to process measurement data with respect to a given task. The model includes a slot building part configured to assign features of an inputted record of measurement data to slots, and a task part configured to process the features that are assigned to each slot separately. The method includes: providing training samples for records of measurement data; producing, from each training sample, by at least one feature extraction layer in the slot building part, at least one feature map; decomposing each feature map into N components according to a predetermined decomposing criterion and/or method; initializing a predetermined number K of slots by assigning, to the slots, features from the feature map, based at least in parts on the N components; and training the slot building part and the task part towards at least one given objective.

## Claims

### Claim at position 1

1-16. (canceled)

### Claim 17

A method for training a machine learning model that is configured to process measurement data with respect to a given task, the model including a slot building part that is configured to assign features of an inputted record of measurement data to a plurality of slots, and a task part that is configured to process the features of the record of measurement data that are assigned to each slot separately, the method comprising the following steps:
providing training samples for records of measurement data;
producing, from each training sample, by at least one feature extraction layer in the slot building part, at least one feature map;
decomposing each feature map into N components according to a predetermined decomposing criterion and/or a predetermined decomposing method;
initializing a predetermined number K of slots by assigning, to the slots, features from the feature map, based at least in parts on the N components; and
training the slot building part and the task part towards at least one given objective.

### Claim 18 (depends on 17)

The method of claim 17, wherein the measurement data includes at least one image, and wherein at least one of the at least one feature extraction layer is a convolutional layer that applies one or more filter kernels to the measurement data in a sliding manner with a predetermined stride.

### Claim 19 (depends on 17)

The method of claim 17, wherein the slots represent object types and/or object instances whose presence the measurement data indicates.

### Claim 20 (depends on 17)

The method of claim 17, wherein both the slot building part and the task part are trained towards one single objective that is measured a result produced by the task part.

### Claim 21 (depends on 17)

The method of claim 17, wherein, during the initializing of the slots, each feature of the at least one feature map is assigned to at most one slot.

### Claim 22 (depends on 17)

The method of claim 17, wherein the predetermined decomposing method includes:
clustering the features of the feature map into a plurality of clusters; and
in response to a feature belonging to a cluster, assigning the feature to a component corresponding to the cluster.

### Claim 23 (depends on 22)

The method of claim 22, wherein the clustering is performed by k-means clustering or mean-shift clustering.

### Claim 24 (depends on 17)

The method of claim 17, wherein the predetermined decomposing method includes:
mapping, by a convolutional layer of the slot building part, each feature to a numeric value; and
assigning the features to components based at least in part on the numeric value.

### Claim 25 (depends on 17)

The method of claim 17, wherein the initializing of the slots includes transforming features that characterize, or belong to, components into features that characterize, or belong to, slots, using at least one trainable transform layer of the slot building part.

### Claim 26 (depends on 25)

The method of claim 25, wherein at least one trainable transform layer is a fully connected layer with a rectified linear unit (ReLU) activation function.

### Claim 27 (depends on 17)

The method of claim 17, wherein the given task includes:
reconstructing original measurement data from the features assigned to the slots; and/or
classifying a type of object whose presence the measurement data indicates; and/or
creating a new record of measurement data that is modified with respect to a particular object whose presence the measurement data indicates; and/or
synthetizing multiple records of measurement data into a new record of measurement data.

### Claim 28 (depends on 17)

The method of claim 17, further comprising:
providing records of measurement data captured by at least one sensor to the trained slot building part;
assigning, by the trained slot building part, features of the records of measurement data to slots; and
processing, by the trained task part, the features assigned to each slot, to obtain a task result with respect to the given task.

### Claim 29 (depends on 28)

The method of claim 28, further comprising:
computing, based on the task result, an actuation signal; and
actuating, using the actuation signal, a vehicle and/or a driving assistance system and/or a surveillance system and/or a quality inspection system and/or a medical imaging system.

### Claim 30

A non-transitory data carrier on which is stored a computer program including machine-readable instructions for training a machine learning model that is configured to process measurement data with respect to a given task, the model including a slot building part that is configured to assign features of an inputted record of measurement data to a plurality of slots, and a task part that is configured to process the features of the record of measurement data that are assigned to each slot separately, the instructions, when executed by one or more computers or compute instances, cause the one or more computers and/or compute instances to perform the following steps:
providing training samples for records of measurement data;
producing, from each training sample, by at least one feature extraction layer in the slot building part, at least one feature map;
decomposing each feature map into N components according to a predetermined decomposing criterion and/or a predetermined decomposing method;
initializing a predetermined number K of slots by assigning, to the slots, features from the feature map, based at least in parts on the N components; and
training the slot building part and the task part towards at least one given objective.

### Claim 31

One or more computers and/or compute instances including a non-transitory data carrier on which is stored a computer program including machine-readable instructions for training a machine learning model that is configured to process measurement data with respect to a given task, the model including a slot building part that is configured to assign features of an inputted record of measurement data to a plurality of slots, and a task part that is configured to process the features of the record of measurement data that are assigned to each slot separately, the instructions, when executed by the one or more computers or compute instances, cause the one or more computers and/or compute instances to perform the following steps:
providing training samples for records of measurement data;
producing, from each training sample, by at least one feature extraction layer in the slot building part, at least one feature map;
decomposing each feature map into N components according to a predetermined decomposing criterion and/or a predetermined decomposing method;
initializing a predetermined number K of slots by assigning, to the slots, features from the feature map, based at least in parts on the N components; and
training the slot building part and the task part towards at least one given objective.
