In tasks like summarization and open-book question answering (QA), Large Language Models (LLMs) often encounter 'contextual hallucination', where they produce irrelevant or incorrect responses despite having access to accurate source information. This typically occurs because these models tend to prioritize self-generated content over the input context, causing them to disregard pertinent details. To address this challenge, a novel method called 'Guided Attention Map Editing' (GAME) is introduced, which dynamically adjusts attention maps to improve contextual relevance. During inference, GAME employs a trained classifier to identify attention maps prone to inducing hallucinations and executes targeted interventions. These interventions, guided by gradient-informed 'edit directions', strategically redistribute attention weights across various heads to effectively reduce hallucination. Comprehensive evaluations on challenging summarization and open-book QA tasks show that GAME consistently reduces hallucinations across a variety of open-source models. Specifically, GAME reduces hallucinations by 10% in the XSum summarization task while achieving a 7X speed-up in computational efficiency compared to the state-of-the-art baselines.
Attention Map Editing, Gradient-guided interventions
Attention-based neural networks
XSum, Open-book QA datasets
Hallucination reduction rate, Computational efficiency
Cloud-based, On-premises
Yes
Yes
Dynamic attention map editing, Gradient-guided interventions
Yes
GPU for training and inference
Windows, Linux, macOS
Compatible with various NLP models
Data privacy and security measures
GDPR compliant
None
No
Available through forums and support channels
Research team from the study
Large-scale datasets
Low latency
Optimized for computational efficiency
Visual representation of attention map adjustments
Ensuring accurate and relevant model outputs
Limited to specific NLP tasks
Technology, Media, Education
Text summarization, Open-book question answering
Tech companies, Educational institutions
Can be integrated with existing NLP pipelines
Highly scalable
Enterprise support available
Available upon request
Web-based interface
Yes
Supports multiple languages
Subscription-based
Yes
Collaborations with NLP research groups
Pending
Compliant with industry standards
1.0
Software as a Service (SaaS)
Yes
RESTful API available
B2B
0.00
USD
Proprietary
01/03/2025
01/03/2025
123-456-7890
Supports integration with popular NLP frameworks
Yes