-
Noisy Zero-Shot Coordination: Breaking The Common Knowledge Assumption In Zero-Shot Coordination Games
Authors:
Usman Anwar,
Ashish Pandian,
Jia Wan,
David Krueger,
Jakob Foerster
Abstract:
Zero-shot coordination (ZSC) is a popular setting for studying the ability of reinforcement learning (RL) agents to coordinate with novel partners. Prior ZSC formulations assume the $\textit{problem setting}$ is common knowledge: each agent knows the underlying Dec-POMDP, knows others have this knowledge, and so on ad infinitum. However, this assumption rarely holds in complex real-world settings,…
▽ More
Zero-shot coordination (ZSC) is a popular setting for studying the ability of reinforcement learning (RL) agents to coordinate with novel partners. Prior ZSC formulations assume the $\textit{problem setting}$ is common knowledge: each agent knows the underlying Dec-POMDP, knows others have this knowledge, and so on ad infinitum. However, this assumption rarely holds in complex real-world settings, which are often difficult to fully and correctly specify. Hence, in settings where this common knowledge assumption is invalid, agents trained using ZSC methods may not be able to coordinate well. To address this limitation, we formulate the $\textit{noisy zero-shot coordination}$ (NZSC) problem. In NZSC, agents observe different noisy versions of the ground truth Dec-POMDP, which are assumed to be distributed according to a fixed noise model. Only the distribution of ground truth Dec-POMDPs and the noise model are common knowledge. We show that a NZSC problem can be reduced to a ZSC problem by designing a meta-Dec-POMDP with an augmented state space consisting of all the ground-truth Dec-POMDPs. For solving NZSC problems, we propose a simple and flexible meta-learning method called NZSC training, in which the agents are trained across a distribution of coordination problems - which they only get to observe noisy versions of. We show that with NZSC training, RL agents can be trained to coordinate well with novel partners even when the (exact) problem setting of the coordination is not common knowledge.
△ Less
Submitted 7 November, 2024;
originally announced November 2024.
-
Code Compass: A Study on the Challenges of Navigating Unfamiliar Codebases
Authors:
Ekansh Agrawal,
Omair Alam,
Chetan Goenka,
Medha Iyer,
Isabela Moise,
Ashish Pandian,
Bren Paul
Abstract:
In our research, we investigate the challenges that software engineers face during program comprehension, particularly when debugging unfamiliar codebases. We propose a novel tool, CodeCompass, to address these issues. Our study highlights a significant gap in current tools and methodologies, especially the difficulty developers encounter in effectively utilizing documentation alongside code explo…
▽ More
In our research, we investigate the challenges that software engineers face during program comprehension, particularly when debugging unfamiliar codebases. We propose a novel tool, CodeCompass, to address these issues. Our study highlights a significant gap in current tools and methodologies, especially the difficulty developers encounter in effectively utilizing documentation alongside code exploration. CodeCompass tackles these challenges by seamlessly integrating documentation within the IDE, offering context-aware suggestions and visualizations that streamline the debugging process. Our formative study demonstrates how effectively the tool reduces the time developers spend navigating documentation, thereby enhancing code comprehension and task completion rates. Future work will focus on automating the process of annotating codebases, creating sandbox tasks, and providing dynamic support. These innovations could potentially transform software development practices by improving the accessibility and efficiency of program comprehension tools.
△ Less
Submitted 10 May, 2024;
originally announced May 2024.
-
Event Coreference Resolution Using Neural Network Classifiers
Authors:
Arun Pandian,
Lamana Mulaffer,
Kemal Oflazer,
Amna AlZeyara
Abstract:
This paper presents a neural network classifier approach to detecting both within- and cross- document event coreference effectively using only event mention based features. Our approach does not (yet) rely on any event argument features such as semantic roles or spatiotemporal arguments. Experimental results on the ECB+ dataset show that our approach produces F1 scores that significantly outperfo…
▽ More
This paper presents a neural network classifier approach to detecting both within- and cross- document event coreference effectively using only event mention based features. Our approach does not (yet) rely on any event argument features such as semantic roles or spatiotemporal arguments. Experimental results on the ECB+ dataset show that our approach produces F1 scores that significantly outperform the state-of-the-art methods for both within-document and cross-document event coreference resolution when we use B3 and CEAFe evaluation measures, but gets worse F1 score with the MUC measure. However, when we use the CoNLL measure, which is the average of these three scores, our approach has slightly better F1 for within- document event coreference resolution but is significantly better for cross-document event coreference resolution.
△ Less
Submitted 9 October, 2018;
originally announced October 2018.
-
A Color Image Digital Watermarking Scheme Based on SOFM
Authors:
J. Anitha,
S. Immanuel Alex Pandian
Abstract:
Digital watermarking technique has been presented and widely researched to solve some important issues in the digital world, such as copyright protection, copy protection and content authentication. Several robust watermarking schemes based on vector quantization (VQ) have been presented. In this paper, we present a new digital image watermarking method based on SOFM vector quantizer for color ima…
▽ More
Digital watermarking technique has been presented and widely researched to solve some important issues in the digital world, such as copyright protection, copy protection and content authentication. Several robust watermarking schemes based on vector quantization (VQ) have been presented. In this paper, we present a new digital image watermarking method based on SOFM vector quantizer for color images. This method utilizes the codebook partition technique in which the watermark bit is embedded into the selected VQ encoded block. The main feature of this scheme is that the watermark exists both in VQ compressed image and in the reconstructed image. The watermark extraction can be performed without the original image. The watermark is hidden inside the compressed image, so much transmission time and storage space can be saved when the compressed data are transmitted over the Internet. Simulation results demonstrate that the proposed method has robustness against various image processing operations without sacrificing compression performance and the computational speed.
△ Less
Submitted 13 January, 2011;
originally announced January 2011.
