LLM Coding Round:

This round involves working with a question-answering model. The first task is to extract relevant information from JSON files in a directory and generate a CSV file for training data. The second task is to identify three hidden bugs that cause errors, which can be found through traceback analysis.

Python Coding Round:

In this round, you were asked to build a graph based on a given Python dictionary and then find the shortest distance between two nodes. The task was framed in the context of Scale.ai's specific scenario. You encountered challenges with indexing while building the graph, which consumed much of your time.

Behavioral Questions (BQ) Round:

This round seems consistent with previous experiences shared in the community. You were asked to discuss the most meaningful project you've worked on, situations that motivate you or demotivate you, how you handle a lack of motivation, your biggest weakness, and an experience where you stepped out of your comfort zone.

Machine Learning Fundamentals Round:

This round started with a discussion of your background, followed by general ML questions (e.g., dealing with overfitting). You chose to focus on Computer Vision (CV) over Natural Language Processing (NLP). You were asked to describe the structure of CNNs and ViTs, their advantages and disadvantages, and considerations for using deep learning models in classification, object detection, and semantic segmentation tasks. Additionally, you were asked about methods to map 3D points to 2D space. Finally, you were given a research paper on masked transformers to summarize its content, innovations, and the model's input and output.

In the context of machine learning, if you are given the task of mapping a collection of 3D points into a 2D space, what methods would you consider using for this transformation? Please explain your choice of methods and the reasoning behind them.