AIAI Seminar-Monday 18 March 2024-Talk by Ameer Saadat-Yazdi, Filip Smola and Yifei Xie

 

Speaker: Ameer Saadat-Yazdi

 

Title: Beyond Recognising Entailment: Formalising Natural Language Inference from an Argumentative Perspective

 

Abstract: In argumentation theory, argument schemes provide a foundation that offers a characterisation of stereotypical patterns of inference. There has been little work done in providing computational approaches to identify these schemes in natural language. Moreover, advancements in recognizing textual entailment lack a standardized definition, which makes it challenging to compare methods trained on different datasets. In this work, we propose a rigorous approach to align entailment recognition with argumentation theory. Wagemans' Periodic Table of Arguments~(PTA), a taxonomy of argument schemes, provides the appropriate framework to unify these two fields. To operationalise the theoretical model, we introduce a tool to assist humans in annotating arguments according to the PTA. Beyond providing insights into non-expert annotator training, we present Kialo-PTA24, the first multi-topic dataset for the PTA. We benchmark the performance of pre-trained language models on various aspects of argument analysis. Our experiments show that the task of argument canonicalisation poses a significant challenge for state-of-the-art models, suggesting an inability to represent argumentative reasoning and a direction for future investigation.

 

 

Speaker: Filip Smola

 

Title: Mechanising Tensors in Isabelle/HOL

 

Abstract: In this talk I will describe our ongoing effort to mechanise tensors in Isabelle/HOL. Tensors are a generalisation of matrices used to represent and analyse structured data. They have been used in many fields, including machine learning. We build on a small existing mechanisation and we both widen the range of defined operations and allow for automated generation of verified code. The focus of my talk will be on how we extend this existing mechanisation to support code generation without interfering with it.

 

 

Speaker: Yifei Xie

 

Title: Leveraging Deep Learning to Tackle Optimization Problems with Constraints

 

Abstract: Optimization problems with hard constraints permeate numerous sectors such as manufacturing, supply chains, and healthcare, where present significant challenges to operational efficiency and decision-making processes. Traditionally, these complex problems not only require experts to formulate them precisely but have also necessitated substantial reliance on heuristic approaches, dependent largely on the domain-specific expertise of professionals in the field.

The advent of AI offers a promising paradigm shift in addressing constrained optimization problems. In this talk, we will explore the transformative potential of AI in solving optimization problems embedded with hard constraints, reducing the dependency on expert knowledge, and paving the way for more scalable, efficient, and universally applicable solutions.