Byunghyun’s Pensées

“Hi I am Byunghyun Kim. ”

I have six years of experience in deep learning for structural health monitoring, participating in 10+ projects in Korea and authoring 10+ papers with 600+ citations. Currently, I'm a postdoctoral scholar at the University of Illinois, mentored by Professor Billie F. Spencer, Jr.

Byunghyun Kim, Ph.D.

Featured Posts

KR

Search

Shadow Mode

The ML system operates alongside the human, mirroring their actions.

Outputs from the ML system are not employed in decision-making at this stage.

Canary Deployment

Initially introduce the system to a minimal portion of the traffic.

Observe the system's performance and incrementally increase the volume of traffic.

Blue-Green Deployment

Traffic is slowly shifted from an old version to a new version.

Levels of Automation

Solely human-operated, Shadow mode, AI-supported, Semi-automated, Completely automated

References

[1] https://www.coursera.org/learn/introduction-to-machine-learning-in-production/lecture/DugTF/deployment-patterns

Deployment Patterns of Machine Learning Model

What is drift in a machine learning project?

In machine learning, drift refers to changes in the distribution of the data.

Concept drift vs Data drift

Concept drift refers to changes in the relationship between the input data and the target variable to be predicted.

Data drift refers to changes in the distribution of the input data.

Sudden shock vs Gradual change

Sudden shock is a change happen in very short period of time. 

Gradual change happens for a long time. 

Reference

[1] https://www.coursera.org/learn/introduction-to-machine-learning-in-production/lecture/k9iID/key-challenges

[2] https://whylabs.ai/blog/posts/data-drift-vs-concept-drift-and-why-monitoring-for-them-is-important

Concept Drift and Data Drift

Completion Certificate for Structuring Machine Learning Projects

This certificate verifies my successful completion of DeepLearning.AI's "Structuring Machine Learning Projects" on Coursera

https://coursera.org/share/42c57a36be592a4e14989a829949d52a

Coursera #3 Structing Machine Learning Project

Completion Certificate for Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization

This certificate verifies my successful completion of DeepLearning.AI's "Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization" on Coursera

https://www.coursera.org/account/accomplishments/verify/GHQCFD2XYJ6L?utm_source=android&utm_medium=certificate&utm_content=cert_image&utm_campaign=sharing_cta&utm_product=course

Coursera #2 Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization

Completion Certificate for Neural Networks and Deep Learning

This certificate verifies my successful completion of DeepLearning.AI's "Neural Networks and Deep Learning" on Coursera

https://coursera.org/share/e7917bfb746fe4defc78f120ff26b007

Coursera #1 Neural Networks and Deep Learning

What is Mean Intersection over Union (mIoU)?

mIoU is one of the representative performance metrics for semantic segmentation.

It is calculated by computing the intersection over union for each segmented class and then taking the average across all classes.

The equation of IoU is defined as below:

{IoU} = \frac{TP}{TP+FP+FN}

Here, TPTPTP represents True positive, FPFPFP represents False Positive, and FNFNFNrepresents False Negative.

In the figure, the green area represents the Ground Truth and the red area represents the Prediction, as indicated by TPTPTP, FPFPFP, and FNFNFN.

Calculating mIoU

Ground Truth and Prediction

Generally, the output of segmentation consists of ground truth (GT) and prediction, as shown in the figure below.

GT and Prediction include integers corresponding to the object class for each pixel.

In the example below, the total number of classes CCC is represented as 5, ranging from 0 to 4.

If implemented as a PyTorch Tensor, it would look like this:

Calculation of Mean Intersection over Union