Now Apple is introducing a no-code AI platform

0


Recently, Apple researchers including CV Krishnakumar Iyer, Feili Hou, Henry Wang, Yonghong Wang, Kay Oh, Swetava Ganguli, Vipul Pandey, Trinity have developed a no-code AI platform for complex spatial data sets.

The platform enables machine learning researchers and non-technical spatial data specialists to experiment with domain-specific signals and datasets to solve various challenges. It adapts complex spatio-temporal data sets to standard deep learning models – in this case convolutional neural networks (CNNs) and formulate different problems in a standard form, e.g. semantic segmentation.

Complete the survey: Using Behavioral Science to Analyze Customer Behavior

“It creates a common vocabulary that leads to better collaboration between domain experts, machine learning researchers, data scientists, and engineers. Currently, the focus is on semantic segmentation, which can easily be extended to other techniques such as classification, regression and instance segmentation, ”the paper says.

Challenges

With the increase in smart devices, a large amount of data is generated and recorded with geo-referenced information. ML techniques have now entered the geospatial field, including hyperspectral image analysis and high resolution satellite image interpretation. However, the use of such solutions is still limited due to specific challenges such as:

  • The processing of large amounts of spatio-temporal information and the application of ML solutions require special skills and therefore have a high barrier to entry, which makes it impossible for non-technical domain specialists to solve problems themselves.
  • The solution is different because residential data is very different from commercial data, resulting in non-standard preprocessing, postprocessing, model deployment, and maintenance workflows.
  • Engineers process data while scientists conduct experiments on different problems and go back and forth a lot. This hinders cooperation.

Trinity faces these challenges:

  • Bring information in different spatio-temporal datasets into a standard format by applying upstream complex data transformations.
  • Standardization of technology to solve different looking problems in order to avoid heterogeneous solutions.
  • Providing an easy-to-use, code-free environment for quick experimentation, lowering the bar for getting started.

It enables rapid prototyping, rapid experimentation, and cuts time to production by standardizing model creation and deployment.

See also

Tech stack

Trinity consists of data pipelines, an experiment management system, a user interface and a containerized deep learning kernel.

  • The platform’s functional memory is managed in S3 (Simple Server Storage). Intermediate data, inputs and processed predictions are stored in a distributed file system (HDFS). Metadata about the experiments, including versions of models, is stored in an instance of a PostgreSQL database that runs on an internal cloud infrastructure.
  • Internal compute clusters that host the GPU and CPU.
  • The training is containerized with Docker and orchestrated by Kubernetes, which runs on the GPU cluster for portability and packaging. Large-scale distributed predictions are performed on CPU clusters orchestrated by YARN.
  • Tensorflow 2.1.0 for training deep learning models. Spark on Yarn for data preprocessing, channel processing, label handling, etc.

The deep learning kernel is the heart of the platform and encapsulates neural network architectures for semantic segmentation and enables model training, evaluation, metrics handling and inference. The kernel is currently implemented in TensorFlow, but can easily be exchanged for other frameworks.


Join our telegram group. Become part of a dedicated online community. Join here.

Subscribe to our newsletter

Get the latest updates and relevant offers by sharing your email.

Kumar Gandharv

Kumar Gandharv

Kumar Gandharv, PGD in English Journalism (IIMC, Delhi), embarks on a journey as a tech journalist at AIM. A keen observer of national and IR-related news. He loves going to the gym. Contact: [email protected]



Source link

Share.

Comments are closed.