International Business Machines Corporation

Honda Elesys Co., Ltd.

Shingo Nagai

Takashi Saito

Naoki Saito

Satoshi Baba

Kazunori Jinno

Hiroshi Oikawa

Yoshinori Masuda

Kohhei Nomura

Takehiro Wakabayashi

Nobuaki Takahashi

Yutaka Kawai

Yohichi Miwa

Fumitomo Ohsawa

Madoka Yamagami

Hiroki Kawasaki

Yiru Lai

Hiroki Kawaski

A training dataset including a plurality of labeled images can be received, each labeled image of the plurality of labeled images including a label represented by a bounding box having a designated class name. Characteristics of each label can be obtained. Statistical information for each class name of the plurality of labeled images can be determined based on the characteristics of each label associated with each respective class name. An outlier test can be performed based on the determined statistical information for a first designated class name to receive at least one outlier label having the first designated class name. The at least one outlier label can be output as a modification candidate for re-labeling.

Image recognition training data quality enhancement

According to one embodiment, a method, computer system, and computer program product for object detection. The embodiment may include receiving an annotated image dataset comprising rectangles which surround objects to be detected and labels which specify a class to which an object belongs. The embodiment may include calculating areas of high and low probability of rectangle distribution for each class of objects within images of the dataset. The embodiment may include applying a correction factor to confidence values of object prediction results, obtained during validation of a trained object detection (OD) model, depending on a class label and a rectangle location of an object prediction result and calculating an accuracy of the trained OD model. The embodiment may include increasing the correction factor and re-calculating the accuracy of the trained OD model with every increase. The embodiment may include selecting an optimal correction factor which yields a highest accuracy.

Object detection considering tendency of object location

An apparatus, a method, and computer program product are provided that tracks data for, and generated by, machine learning for accurate and precise deletion. The method includes receiving a dataset for use in training a machine learning model and registering a file from the dataset into a reference table, wherein the file is designated for monitoring. The file designation can indicate that the file is confidential and requires deletion upon completion of training of the machine learning model and project. The method also includes monitoring the file for an event that accesses the file, detecting a read access event occurring on the file, and determining a creation of a derivative file generated as a result of the read access event. The method further includes registering the derivative file into the reference table and indicating an association between the derivative file and the file in the reference table.

Tracking techniques for generated data

Disclosed are techniques for augmenting video datasets for training machine learning algorithms with additional video datasets that are cropped copies of the video datasets. Frames of a received video dataset are divided into a plurality of subframes. For each subframe, a count is tallied corresponding to the cumulative number of pixels changed across the frames of the received video. Counts are compared to determine which subframe includes the most changed pixels across the frames of the video dataset, which is selected as a cropping candidate. The cropping candidate is used to generate copies of the video dataset that are cropped to include at least the cropping candidate and exclude at least some of the remaining portions of each frame of the video dataset that are outside of the cropping candidate. In some embodiments, boundaries of cropping candidates are transformed to generate a plurality of cropped variations of the video dataset.

Augmentation of video datasets for machine learning training

A computer-implemented method, a computer program product, and a computer system for efficient use of computing resources in two stage training of a deep learning model. A computer executes a first first-stage training job to train a deep learning model. The computer finishes the first first-stage training job by using early stopping and then registers a first second-stage training job to train a deep learning model that has been trained in the first first-stage training job. The computer executes the first second-stage training job with a small number of epochs. The computer interrupts the first second-stage training job and executes a second first-stage training job, in response to receiving a registration of the second first-stage training job. The computer interrupts the first second-stage training job and executes a second second-stage training job that has a higher priority, in response to receiving a registration of the second second-stage training job.

Two stage training to obtain a best deep learning model with efficient use of computing resources

A method for balancing load across multiple file servers is disclosed. In one embodiment, such a method includes monitoring load experienced by multiple file servers arranged in an active-active configuration. The method receives, from a client node, a request for an address associated with one of the file servers. The method determines a particular file server of the file servers that, if assigned to the client node, would reduce load imbalance between the file servers. In certain embodiments the particular file server is the file server experiencing the least load. The method returns, to the client node in response to the request, an address associated with the particular file server, thereby enabling the client node to mount the particular file server to access files thereon. A corresponding system and computer program product are also disclosed.

File server load balancing

A method for generating training data for object detection is provided. The method includes dividing an image into a plurality of sliced images. The image includes meta information for an object in the image, the meta information including bounding box information associated with a bounding box surrounding the object, and marker information for markers associated with the object, the markers being located within the bounding box. The method includes designating at least one sliced object image from among the plurality of sliced images, at least a portion of the bounding box overlapping with each of the respective sliced object images. The method also includes designating at least one training data image from among the designated sliced object images based on the marker information. The method also includes generating the training data from the designated training data images.

Generating training data for object detection

A computer initiates a fault prevention shell. The computer protects a plurality of the computer's files in a first storage area. The computer carries out a command entered by a user into the fault prevention shell, wherein the command targets one or more of the plurality of the computer's files in the first storage area, and wherein the command is carried out on a copy of the one or more of the plurality of the computer's files in a second storage area. The computer prompts a commit by the user to perform the command on the one or more of the plurality of the computer's files in the first storage area. The computer processes a user response to the prompt. The computer updates one or more command lists with the command.

Fault prevention shell for preventing system disruption

Embodiments of the present invention are directed to a computer-implemented method for simulating a plurality of electronic control units ('ECU's) in communication over a simulated bus. The method includes simulating an operation of a first ECU and an operation of a second ECU and performing arbitration at a packet-level granularity at a packet transmission start point with respect to a first packet sent to the simulated bus by the first ECU and a second packet sent to the simulated bus by the second ECU. The method identifies an initially winning ECU in the arbitration and a zone from the packet transmission start point to a bit where the initially winning ECU is determined to win based on the arbitration and continues the simulation of the operation of the first ECU and the operation of the second ECU to the end of the zone.

Accurate can-based distributed control system simulations

There is provided a system including a cluster, a clusterand an image server. A hostconstituting the clusterhas a root file systemused to operate the hostin a local disk. The image serverhas a root file systemused to operate the host, and the hostvirtually constructs a root file systemby mounting the root file system. When the root file systemis varied, the hostsynchronizes the root file systemwith the root file system. The same applies to hostconstituting the cluster

System and method for managing root file system

In an occupant detecting apparatus for detecting an occupant seated on a passenger seat of a vehicle with an airbag for the occupant, a load sensor is provided in a bottom part of the seat. A plurality of first electric field sensors are provided in the bottom part of the seat, and a plurality of second electric field sensors are provided in a rear part of the seat. An airbag inflating permission control unit permits inflation of the airbag in accordance with output signals of the load sensor and the first and second electric field sensors.

Growing community of inventors

Tokyo, Japan

Shingo Nagai