1. Explore ImageNet. ImageNet sample images, Kaggle ImageNet Mini 1000, What surprises you about this data set? What questions do you have? Thinking back to last week’s assignment, can you think of any ethical considerations around how this data was collected Are there privacy considerations with the data?

I am surprised that the images in the ImageNet dataset are pretty generic and random. For the displayed images, I don't see very close-up shots of the object that give unique identification to each class. I am wondering how the algorithm can differentiate between the subtle differences between animals within a species, for example the different types of spiders.

For ethical considerations, the dataset should list all the sources of the images that are collected and used to develop ImageNet. Also, how do we define the ownership of these images? For privacy considerations, there are images that include people. Their faces should not be scanned or studied for the development of this model. I am not sure whether unrelated objects could be ignored or omitted for machine learning.

1. Using the ml5.js examples above, try running image classification on a variety of images. Pick at least 10 objects in your room. How many of these does it recognize? What other aspects of the image affect the classification, including but not limited to position, scale, lighting, etc.

Not many of my objects are recognized correctly. For example, a scissor is classified as spatula, my headphones are recognized as facial masks, and my vitamins bottle are recognized as hair spray (I don't know why hair spray comes up so often; basically, every bottle-like object is recognized as hair spray).

I also tried to pull up an image on my phone and see what the image classification would recognize it as. As long as I put my phone close enough, the image classification can recognize the “fake” object. In a sense, it is doing the right thing because it gives the right classification. But I start to think about how this model can be trained to be smarter. What if the model is able to tell that is an image or a real object?