KARI-AMBER-AI

Training neural networks with amber fossil morphological traits with images, and text.

Intersection of AI and Burmite Kachin Amber

Since late 2023 the Kachin Amber Research Institute was approached by physics and mathematicians, data scientists associated with emerging artificial intelligence research in the city of San Diego, California.

The widespread adoption of Artificial Intelligence tools is gripping many industries. The Kachin Amber Research Institute recently launched an internal project to collaborate with AI researchers in America to fund and begin collating a large repository of amber images taken by internal kachin amber research specialists. These images alone are insufficient, and must be combined with thorough classification and entomological/paleontological insight.

The essential goal and drive is to combine entomological and taxonomical classifications with accompanying images of amber inclusions from the Hukawng Valley in Kachin State. Many pre-existing artificial intelligence models are not trained to much degree of accuracy to identify prehistoric organisms.

Developing KARI-AMBER-AI

After securing a research grant our board began the first steps to collaboration with San Diego Artificial Intelligence specialists with background in training in maths, physics and chemistry from the prestigious University of San Diego. Data collection and cleaning is underway. Using OpenAI's contrasive language image pre-training, numerous neural networks including ResNet architecture for images (and others for text), tokenization and tensor creation/node comparisons are mapped out for Q3 2024 with the KARI-AMBER-AI team.

The neural network is being trained with high quality amber inclusion photographs of amber from the hukawng valley. Collection of various datasets are a work in progress; however the exact datasets and their labeling for the neural network are not public information. These image datasets are not publicly available and their eventual collation, along with millions of other images publicly available on the internet, will begin to train the model to make discerning observations.

Technologies:
KARI-AMBER-AI is essentially multidisciplinary. The team seeks to bridge the gap between traditional amber research methods of naming, describing organisms in Cretaceous Burmite amber, with knowledge of data collection, storage, dataset cleaning and parsing, and ultimately to machine learning.

The technology stack is evolving as the project matures, but here are some components KARI-AMBER-AI is powered on:
- OPENAI-CLIP
- Terraform
- NVIDIA T4 distributed GPU's reserved with virtual machines in Sao Paulo, Brazil.
- Ubuntu
- MongoDB
- NodeJS
- Bash Shell Programming language
- awk & regex filtration
- ResNet

Team:
The team includes specialists abroad and within Myanmar and Kachin State.

Local:
- Myint Thwe, Biochemist
- John La Aung, KARI Macropaleophotography
International:
- Ian Abramowitz, Lead Engineer, CLIP Taskforce KARI-AMBER-AI, B.S. Biochemistry & Chemistry UCSD

AI HIRING - work with us

A placement at the Kachin Amber Research Institute means contributing to Myanmar's place in the world.

March 2024 - [(REMOTE)] AI model trainer (part-time placement)

Description: Must hold a bachelors degree (minimum) in the sciences with a proven background in the field of entomology from a reputable scientific institution.