Aditya K Surikuchi

I am currently a PhD candidate at the University of Amsterdam, Netherlands. Multimodality and interpretability in machine learning are my primary areas of interest.
I obtained my Master's in Machine learning and Artificial intelligence from Aalto University, Finland.

Before pursuing Master's, I worked as a Software Development Engineer for 3 years at CommerceIQ.AI and ACS solutions.
I obtained my Bachelor's in Computer Science and Engineering from Amrita University, India.

The following sections link to my academic work. Details related to other projects and industry experience are available through these sources:

LinkedIn  |  Google Scholar  |  a.k.surikuchi@uva.nl

profile photo

Research

Character-Centric Storytelling
Aditya K Surikuchi, Jorma Laaksonen
ACL StoryNLP, 2019

Sequential vision-to-language or visual storytelling has recently been one of the areas of focus in computer vision and language modeling domains. Though existing models generate narratives that read subjectively well, there could be cases when these models miss out on generating stories that account for and address all prospective human and animal characters in the image sequences. Considering this scenario, we propose a model that implicitly learns relationships between provided characters and thereby generates stories with respective characters in scope.

DeepCaption
Jorma Laaksonen, Mats Sjöberg, Aditya K Surikuchi, Arturs Polis
PyTorch-based framework, 2018
code

Image captioning is a process of generating textual descriptions for images. Deep captioning is a framework built on the widely adapted encoder-decoder model of automated image captioning, with additional features such as scheduled sampling and attention mechanism.


Master's thesis
Visual Storytelling
Aditya K Surikuchi, Jorma Laaksonen (supervisor)
2018-19

Visual storytelling is a sequential vision-to-language task of automatically generating contextual and subjective textual narratives for image sequences or video frames. Through my master’s thesis, I surveyed the existing methods, implemented them, and proposed improvements. Additionally, my thesis explored the sub-topic of character relationships between textual and visual modalities.


Other selected projects
Evaluation mechanisms for Generative models

2019

Evaluating and assessing generative models is challenging, primarily owing to the implicit lack of ground truths. Additionally, the inherent nature of subjectivity associated with the quality of generated samples makes evaluation difficult. Through this project, I compared and discussed the implications of various methodologies ranging from employing an external classifier model for automatic scoring to using Bayesian neural networks for quantifying uncertainty.

Neural Language Models

2018

Implemented several RNN (recurrent neural network) based language models with different parameters and applied interpolation to achieve a single standalone model. Trained and tested the model using the English Gigaword corpus and intrinsically evaluated using perplexity. This project was a part of my master’s coursework in speech recognition.

Visualization of Adversarial Loss

2018

Ensemble adversarial training claimed to improve the robustness of deep learning models against adversarial attacks without merely masking the gradients of the model parameters. To validate the claims, I modeled an image classifier using the proposed training procedure, visualized and analyzed the resulting loss landscapes.



©Thanks a lot for this template Jon!