AI Researcher
Personal Philosophy:
The more one moves away from one's own nature, the more one must justify existing.
It is a humbling starting point to acknowledge that in life you don't know why you consciously exist (exempting a material explanation of being such as ancestry). In the absence of a reason, one requires meaning and purpose to fill its void. Without over thinking it, my most meaningful experiences have been a clear articulation of impressions of the present.
Modern Marvels:
Neural network feature visualization by optimization: what features of an image biases classification? These authors fix the weights of a trained network as an optimization objective to search over raw input images. The groovy images on the left are the result, revealing insights into what features the network is sensitive to for correct and even... incorrect (false positive) classification!
Visualization of a robot's "artificial brain" for solving a Rubik's cube: each row corresponds to the activation values of a policy network's LSTM cell state (1024 in total) over 1500 time steps. Dashed lines mark a successful cube re-orientation, solid lines mark a successful face rotation.
Writings coming soon...