Example of my Cover Letter for AI Residency application. Supporting document to this video.
- DISCLAIMER:
- This is not example of a PERFECT cover letter, is example of MY cover letter.
- It's quite old (2018), I would write it very differently today. Cover letter
Oleksii Sidorov
Why I consider myself a perfect candidate for Facebook AI Residency, in a nutshell:
- strong theoretical background (advanced Physics and Math)
- experience in Deep Learning and Computer Vision
- publication history
- codding skills in MatLab, Python, C++
- outstanding academic performance, fast learning ability
- experience of working in multi-cultural team with frequent relocations
- fresh graduate in 2019, open to the next steps (22 yo)
And now the same but in details:
Currently, I am finishing Master program COSI (Color in Science and Industry). This is a joint program with academic semesters in France, Spain, and Norway. It involves the strongest students from all over the world (because of Erasmus scholarship) and guarantees the highest level of education and an experience of work in a multicultural team. It covers many different fields of study, such as Human Vision, Computer Vision, Image Processing, Data Science, Optics, etc. I got here because of my Bachelor degree in Physics, particularly Optics and Photonics, but eventually discovered that new domain is much more interesting than what I had studied before.
The previous degree provided me with a strong theoretical basis in calculus, scientific programming, linear algebra, and optics which are surely important for any Computer Vision research. However, COSI program did something greater – it uncovered my potential and helped me to find my field of interest. Probably the clearest evidence of this is that at the beginning of the program I knew nothing about computer vision, machine learning and vision science, but just in one academic year, I had seven papers accepted to publication, in six of which I am the main author. Besides that, all this time I held the first position in group ranking, between students who were selected by their outstanding performance, and who were much more experienced in computer science and also older than me. This is a strong proof of my skills even for myself – currently I have much more confidence and brevity to discover new areas and solve unfamiliar problems than I had before this program.
My deep learning experience includes usage of various architectures, namely:
- basic Multilayer Perceptrons for regression and classification
- conventional deep CNNs for regression, classification, and feature visualization (AlexNet, VGG##, ResNet-##, Inception-v#, GoogLeNet)
- segmentation algorithms (SegNet, U-Net, FCN, DeepLab)
- Generative Adversarial Networks (pix2pix, CycleGAN, AttGAN, StarGAN, VAE-GAN) The experiments were done using either PyTorch, Tensorflow or Matlab NN Toolbox.
Talking about research in general, my expertise lies all around Computer Vision: vision, machine learning, image processing, optics. Logically, I consider Computer Vision to be the most suitable direction for my further development. Importance of it in the modern world requires no explanation. If we want artificial machines to be autonomous – they need to see. Likewise to a human who lost their sight, we can imagine how dramatically performance depends on the perception of visual information. Nowadays, despite significant progress, the biggest obstacle is the fragmentation of the algorithms by niche tasks. In the future, I want to pursue the idea of unification of CV algorithms. The ideal CV model, in my opinion, should be universal. Not only it will make model easier to use but will also allow to combine knowledge from different domains, just as the human brain does it (and as it is implied in the concept of transfer learning).
As an example of completed research task I want to discuss image memorability. This is quite a novel topic at the intersection of computer vision and psychology. What exactly makes image memorable is still unknown. The semantic analysis shows that images which contain people are usually the most memorable, while natural scenes – are the least memorable. However, it is useless for controlled memorability modification, simply because we cannot add random people on every image. The influence of visual features is rather complex and is usually covered by extraction of feature descriptors and feeding them to machine learning algorithms. However, such format is not explanatory for humans, that makes reverse-engineering impossible. Thus, despite of numerous studies, there are still no clues how to change the memorability. I propose solution of this problem in my recent paper, where I use GAN designed for attribute editing to perform unsupervised learning, and in result generate different versions of an input image with different levels of memorability. There were also other obstacles on the way, such as the creation of a large dataset, but I encourage to check it out in the paper directly. Just to conclude, produced memorability change reached values up to 33%, and what is the most important – generated data samples unachievable in real-life (modification of memorability while leaving other attributes unchanged) which allows to use it for data mining in further research.
Apart from that, I am also quite excited about computational photography, visual computing and image generation. Text to image generation, visual reasoning, answer questioning – all these will lead human-computer interaction for many years ahead. And algorithms by Facebook demonstrate very prominent results in this domain. For me personally, it also would be very interesting to extend my image processing experience to natural language processing.
Participation in AI Residency program, to my view, is exactly what I am prepared for. I want to fulfill my potential, and I am afraid to waste it. I wrote the same when applying for Erasmus scholarship, and I fully met the expectations – become the 1st in the group, produced research output, and gained an excellent reputation in different universities. Now I want to develop this potential. Being surrounded by the atmosphere of innovations, being taught by the best experts in the field, being supported and motivated – this is what I need. Moreover, during this program, I hope to extend my view to real industrial research, since for now my experience is limited to the academic environment, such as working in research labs and teaching the bachelors. Also, I never considered myself a software engineer, but now I feel a strong need to enhance my coding skills, and algorithms knowledge, that I am sure will be done during the Residency.
I have great respect for AI research in Facebook. And I want to contribute to it, creating meaningful products which will change the world.