Bridging the Gap: Exploring the Capabilities of Bridge-Architectures for Complex Visual Reasoning Tasks

In recent times there has been a surge of multi-modal architectures based on Large Language Models, which leverage the zero shot generation capabilities of LLMs and project image embeddings into the text space and then use the auto-regressive capacity to solve tasks such as VQA, captioning, and image retrieval. We name these architectures as "bridge-architectures" as they project from the image space to the text space. These models deviate from the traditional recipe of training transformer based multi-modal models, which involve using large-scale pre-training and complex multi-modal interactions through co or cross attention. However, the capabilities of bridge architectures have not been tested on complex visual reasoning tasks which require fine grained analysis about the image. In this project, we investigate the performance of these bridge-architectures on the NLVR2 dataset, and compare it to state-of-the-art transformer based architectures. We first extend the traditional bridge architectures for the NLVR2 dataset, by adding object level features to faciliate fine-grained object reasoning. Our analysis shows that adding object level features to bridge architectures does not help, and that pre-training on multi-modal data is key for good performance on complex reasoning tasks such as NLVR2. We also demonstrate some initial results on a recently bridge-architecture, LLaVA, in the zero shot setting and analyze its performance.

Exploring the Role of the Bottleneck in Slot-Based Models Through Covariance Regularization

In this project we attempt to make slot-based models with an image reconstruction objective competitive with those that use a feature reconstruction objective on real world datasets. We propose a loss-based approach to constricting the bottleneck of slot-based models, allowing larger-capacity encoder networks to be used with Slot Attention without producing degenerate stripe-shaped masks. We find that our proposed method offers an improvement over the baseline Slot Attention model but does not reach the performance of \dinosaur on the COCO2017 dataset. Throughout this project, we confirm the superiority of a feature reconstruction objective over an image reconstruction objective and explore the role of the architectural bottleneck in slot-based models.

A Multi-Objective approach to the Electric Vehicle Routing Problem

The electric vehicle routing problem (EVRP) has garnered great interest from researchers and industrialists in an attempt to move from fuel-based vehicles to healthier and more efficient electric vehicles (EVs). While it seems that the EVRP should not be much different from traditional vehicle routing problems (VRPs), challenges like limited cruising time, long charging times, and limited availability of charging facilities for electric vehicles makes all the difference. Previous works target logistics and delivery-related solutions wherein a homogeneous fleet of commercial EVs have to return to the initial point after making multiple stops. On the opposing front, we solve a personal electric vehicle routing problem and provide an optimal route for a single vehicle in a long origin-destination (OD) trip. We perform multi-objective optimization - minimizing the total trip time and the cumulative cost of charging. In addition, we incorporate external and real-life elements like traffic at charging stations, detour distances for reaching a charging station, and variable costs of electricity at different charging stations into the problem formulation. In particular, we define a multi-objective mixed integer non-linear programming (MINLP) problem and obtain a feasible solution using the $\epsilon$-constraint algorithm. We further implement meta-heuristic techniques such as Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) to obtain the most optimal route and hence, the objective values. The experiment is carried out for multiple self-generated data instances and the results are thereby compared.

MAP-Elites based Hyper-Heuristic for the Resource Constrained Project Scheduling Problem

The resource constrained project scheduling problem (RCPSP) is an NP-Hard combinatorial optimization problem. The objective of RCPSP is to schedule a set of activities without violating any activity precedence or resource constraints. In recent years researchers have moved away from complex solution methodologies, such as meta heuristics and exact mathematical approaches, towards more simple intuitive solutions like priority rules. This often involves using a genetic programming based hyper-heuristic (GPHH) to discover new priority rules which can be applied to new unseen cases. A common problem affecting GPHH is diversity in evolution which often leads to poor quality output. In this paper, we present a MAP-Elites based hyper-heuristic (MEHH) for the automated discovery of efficient priority rules for RCPSP. MAP-Elites uses a quality diversity based approach which explicitly maintains an archive of diverse solutions characterised along multiple feature dimensions. In order to demonstrate the benefits of our proposed hyper-heuristic, we compare the overall performance against a traditional GPHH and priority rules proposed by human experts. Our results indicate strong improvements in both diversity and performance. In particular we see major improvements for larger instances which have been under-studied in the existing literature.