KARL: Knowledge-Aware Reasoning Memory Modeling with Reinforcement Learning of Vector Space
Founded in Atkinson-Shiffrin Memory Model's three-stage theory, the cognitive process of answering a question with stored knowledge can be seen as a reasoning process that goes from the external sensory memory via short-term knowledge-aware reasoning towards the internal knowledge storage. While in the machine, this process can be interpreted as a pipeline from encoding the question’s information, through decoding the reasoning query of triples, towards the knowledgebase. How to encode and decode the varying semantic information of question into accurate triple query, and how to adjust the query generation with an evolving knowledgebase, are two inevitable problems in this cognitive process. Our model KARL, provides a solution by designing the three memory spaces as, an encoder to handle the language modeling, a decoder for query generation, and a self-calibration with reinforcement learning of the knowledge representation vector space. We evaluate the model's reasoning ability in knowledge-based question answering against the Question Answering over Linked Data (QALD) benchmark, and achieve significant improvements in answers accuracy as compared to other neural models.
Code (0)
등록된 구현이 없습니다.
Tasks
DecoderLanguage ModelingLanguage ModellingQuestion Answeringreinforcement-learningReinforcement Learning (RL)Similar Papers 제목 키워드 기반
KARL-Trans-NER: Knowledge Aware Representation Learning for Named Entity Recognition using Transformers
The inception of modeling contextual information using models such as BERT, ELMo, and Flair has significantly improved representation learning for words. It has also given SOTA results in almost every NLP task - Machine …
Machine Translationnamed-entity-recognitionNamed Entity RecognitionNamed Entity Recognition (NER)+5KARL: Knowledge-Aware Retrieval and Representations aid Retention and Learning in Students
Flashcard schedulers rely on 1) student models to predict the flashcards a student knows; and 2) teaching policies to pick which cards to show next via these predictions. Prior student models, however, just use study dat…
Knowledge TracingRetrievalSchedulingKARL: Mitigating Hallucinations in LLMs via Knowledge-Boundary-Aware Reinforcement Learning
Enabling large language models (LLMs) to appropriately abstain from answering questions beyond their knowledge is crucial for mitigating hallucinations. While existing reinforcement learning methods foster autonomous abs…
Reinforcement LearningKARL: Knowledge Agents via Reinforcement Learning
We present a system for training enterprise search agents via reinforcement learning that achieves state-of-the-art performance across a diverse suite of hard-to-verify agentic search tasks. Our work makes four core cont…
Reinforcement Learningkarl. -- A Research Vehicle for Automated and Connected Driving
As highly automated driving is transitioning from single-vehicle closed-access testing to commercial deployments of public ride-hailing in selected areas (e.g., Waymo), automated driving and connected cooperative intelli…