Assisted Inverse Reinforcement Learning

We study the problem of inverse reinforcement learning (IRL) with the added twist that the learner is assisted by a helpful teacher. More formally, we tackle the following question: How could a teacher provide an informative sequence of demonstrations to an IRL agent to speed up the learning process? We prove rigorous convergence guarantees of a new iterative teaching algorithm that adaptively chooses demonstrations based on the learner’s current performance. Extensive experiments with a car driving simulator environment show that the learning progress can be speeded up drastically as compared to an uninformative teacher.

P. Kamalaruban, R. Devidze, T. Yeo, T. Mittal^*, V. Cevher, A. Singla

Pictionary-style Word Guessing on Hand-drawn Object Sketches: Dataset, Analysis and Deep Network Models

The ability of intelligent agents to play games in human-like fashion is popularly considered a benchmark of progress in Artificial Intelligence. In our work, we introduce the first computational model aimed at Pictionary, the popular word-guessing social game. We first introduce Sketch-QA, a guessing task. We analyze the resulting dataset and present many interesting findings therein. To mimic Pictionary-style guessing, we propose a deep neural model which generates guess-words in response to temporally evolving human-drawn object sketches. Our model even makes human-like mistakes while guessing, thus amplifying the human mimicry factor. We evaluate our model on the large-scale guess-word dataset generated via Sketch-QA task and compare with various baselines.

Ravi Kiran Sarvadevbatla, Shiv Surya, Trisha Mittal^*, Venkatesh Babu Radhakrishnan

A Logo-Based Approach for Recognising Multiple Products on a Shelf

This paper addresses detecting, localizing and recognizing various grocery products in retail store images. Our object recognition algorithm achieves this goal using just one image per product for training, assuming that the category of the products (like cereals, rice, etc.) is known. This algorithm uses logo detection as a precursor to product recognition. So, the first step involves detecting and classifying products, at a broader level, based on their brands. The second step is the finer classification step for recognizing and localizing the exact product label, which involves using colour information. This hierarchical approach limits the confusion in classifying similar looking products and outperforms product recognition that was implemented without logo detection.

Trisha Mittal^*, B. Laasya, and J. Dinesh Babu

CMSC 250

Course Instructor: Prof. Clyde Kruskal

CMSC 250

Course Instructor: Jason Filippou

Visiting Research Fellow at Max-Planck Institute for Software Systems, Saarbrucken, Germany

Interned under Prof. Adish Singla and worked on problems at the intersection of Inverse Reinforcement Learning and Machine Teaching.

Intern at Video Analytics Lab, Indian Institute of Science, Bangalore, India.

Guided by Prof. Venkatesh Babu in an attempt to train a RL agent to be able to draw sketches.

Intern at Video Analytics Lab, Indian Institute of Science, Bangalore, India.

Guided by Prof. Venkatesh Babu and worked with Ravikiran Sarvadevbatla and Shiv Surya to develop computational models for Pictionary-style word guessing.