I am Roshanak Mirzaee, a research scientist with expertise in natural language processing and symbolic reasoning. My journey into programming and artificial intelligence began at the age of 14 when I joined the Robotic team at my school. I graduated with a Bachelor's degree in Computer Science and Engineering from Amirkabir University of Technology. Subsequently, I earned a Master's degree from the University of Tehran, securing the 3rd highest academic standing in the same field.
In 2019, I started my Ph.D. program. During my doctorate studies, my focus was on enhancing language models' spatial reasoning and comprehension capabilities. We proposed several benchmarks for both evaluation and transfer learning in language models. Despite the advancements in LLMs, we identified these models' poor complex reasoning capability and formulated methodologies to mitigate this challenge. We also provide models that solve spatial reasoning problems using models with disentangled extraction and reasoning modules. During my Ph.D., I was fortunate to participate in two summer internships as a Research Scientist Intern at Bosch-LLC and Qualtrics-XM.
My research interests span various topics, including natural language processing, neuro-symbolic reasoning, probabilistic reasoning, multi-hop reasoning, spatial commonsense reasoning, information extraction, neural model interpretability, and Large Language Models.
Computer Science and Engineering (NLP)
Michigan State University (USA)
GPA: 4
Research Assistant: Spatial Language Understanding and extraction, Spatial Reasoning, Symbolic Reasoning, Question Answering, Pretrained Language Models, Transfer Learning, Large Language Models.
Teacher Assistant: Data structure and Algorithm, Introduction to Python
Computer Engineering, Algorithms & Computation
Tehran University (Iran)
GPA: 4
A Case-based Reasoning Approach for recommender system of interior design: Improve CBR by compositional adaptation and fuzzy ontology on semantics relationships of objects to recommend interior design sets.
Computer Science
Amirkabir University Of Technology (Iran)
GPA: 3.7
Research Assistant: Spatial Language Understanding and extraction, Spatial Reasoning, Symbolic Reasoning, Question Answering, Pretrained Language Models, Transfer Learning, Large Language Models.
Teacher Assistant: Data structure and Algorithm, Introduction to Python
Improve the generalizability of models considering the result of interpretation
Method and Metrics for Models Interpretation: Define a saliency score to check the influence of highlighted keywords extracted by saliency map methods (Using AllenAI packages for interpretation .)
Augmenting Language Models with Spatial CommonSense Through Synthetic Question Answering
Developing Front-end of web-pages.
Teaching web development in Vestacamp.
Spatial reasoning over text is challenging as the models not only need to extract the direct spatial information from the text but also reason over those and infer implicit spatial relations. Recent studies highlight the struggles even large-scale language models encounter when it comes to performing spatial reasoning over text. In this paper, we explore the potential benefits of disentangling the processes of information extraction and reasoning to address this challenge. We design various models that disentangle extraction and reasoning~(either symbolic or neural) and compare them with pretrained language model baselines which have state-of-the-art results. Our experimental results consistently demonstrate the efficacy of disentangling, showcasing its ability to enhance models' generalizability within realistic data domains.
Spatial reasoning over text is challenging as the models need to extract the direct spatial information from the text, reason over those, and infer implicit spatial relations. Recent studies highlight the struggles even large-scale language models encounter when it comes to performing spatial reasoning over text. In this paper, we explore the potential benefits of disentangling the processes of information extraction and reasoning in models to address this challenge. To explore this, we devise various models that disentangle extraction and reasoning (either symbolic or neural) and compare them with SOTA baselines with no explicit design for these parts. Our experimental results consistently demonstrate the efficacy of disentangling, showcasing its ability to enhance models’ generalizability within realistic data domains.
Recent research has shown that integrating domain knowledge into deep learning architectures is effective -- it helps reduce the amount of required data, improves the accuracy of the models' decisions, and improves the interpretability of models. However, the research community is missing a convened benchmark for systematically evaluating knowledge integration methods. In this work, we create a benchmark that is a collection of nine tasks in the domains of natural language processing and computer vision. In all cases, we model external knowledge as constraints, specify the sources of the constraints for each task, and implement various models that use these constraints. We report the results of these models using a new set of extended evaluation criteria in addition to the task performances for a more in-depth analysis. This effort provides a framework for a more comprehensive and systematic comparison of constraint integration techniques and for identifying related research challenges. It will facilitate further research for alleviating some problems of state-of-the-art neural models.
Recent research shows synthetic data as a source of supervision helps pretrained language models (PLM) transfer learning to new target tasks/domains. However, this idea is less explored for spatial language. We provide two new data resources on multiple spatial language processing tasks. The first dataset is synthesized for transfer learning on spatial question answering (SQA) and spatial role labeling (SpRL). Compared to previous SQA datasets, we include a larger variety of spatial relation types and spatial expressions. Our data generation process is easily extendable with new spatial expression lexicons. The second one is a real-world SQA dataset with human-generated questions built on an existing corpus with SPRL annotations. This dataset can be used to evaluate spatial language processing models in realistic situations. We show pretraining with automatically generated data significantly improves the SOTA results on several SQA and SPRL benchmarks, particularly when the training data in the target domain is small.
In this chapter, we provide a retrospective analysis of our recent work, wherein we develop methods for integrating neural language models and knowledge graphs. We characterize the situations in which this combination is most appropriate, and we offer quantitative and qualitative evaluation of these neuro-symbolic modeling strategies on a variety of commonsense question answering benchmark datasets. Our overarching goal is to illustrate how suitable combinations of learning mechanisms and knowledge can enable domain generalizability and robustness in downstream tasks. The chapter is structured as follows: section 13.2 describes some limitations observed when state-of-the-art language models are used in commonsense question answering tasks; in section 13.3 we discuss the main findings of our empirical investigations in discriminative tasks, focused on knowledge-injection and pre-training/fine-tuning methods (13.3. 3-13.3. 4), and we present novel ideas about improving reasoning capabilities of neuro-symbolic systems, laying down the path towards generalizability (13.3. 5).
This paper proposes a question-answering (QA) benchmark for spatial reasoning on natural language text which contains more realistic spatial phenomena not covered by prior work and is challenging for state-of-the-art language models (LM). We propose a distant supervision method to improve on this task. Specifically, we design grammar and reasoning rules to automatically generate a spatial description of visual scenes and corresponding QA pairs. Experiments show that further pretraining LMs on these automatically generated data significantly improves LMs' capability on spatial understanding, which in turn helps to better solve two external datasets, bAbI, and boolQ. We hope that this work can foster investigations into more sophisticated models for spatial reasoning over text.
The Best Poster of Michigan AI Symposium 2020. Best poster award Grad symposium MSU 2022
We propose a novel alignment mechanism to deal with procedural reasoning on a newly released multimodal QA dataset, named RecipeQA. Our model is solving the textual cloze task which is a reading comprehension on a recipe containing images and instructions. We exploit the power of attention networks, cross-modal representations, and a latent alignment space between instructions and candidate answers to solve the problem. We introduce constrained max-pooling which refines the max-pooling operation on the alignment matrix to impose disjoint constraints among the outputs of the model. Our evaluation result indicates a 19\% improvement over the baselines.
Using probabilistic logical programming to solve the spatial logical reasoning and spatial commonsense reasoning in Large Language Models.
Evaluate LLMs such as GPT3.5, GPT4 and PaLM2 on Spatial reasoning. Providing A disentangled approach to use LLMs as the extraction modules and use symbolic reasoning to do the reasoning steps.
Propose a model including spatial information extraction and spatial reasoner modules for spatial question answering.
Implement coreference resolution model including plural antecedents.
Define a saliency score to check the influence of highlighted keywords extracted by saliency map methods (Use AllenAI packages for interpretation .)
Use Context-free grammar, context-sensitive rules, and spatial rules to generate automatic text and questions. Propose different experiments to evaluate Language models’ spatial reasoning capability using the generated distant supervision in transfer learning.
We have employed case-based reasoning in a recommender system, which proposes optimal interior design solutions for a scanned environment. Specifically, semantic relations were utilized for superior case (information) retrieval, and compositional adaptation was employed to generate the most appropriate recommendations. When an issue arises, the system identifies analogous cases based on the semantic similarity between previously utilized objects and recommends objects that fulfill user requirements. Additionally, machine learning techniques were integrated to enhance the overall process, which facilitate the system's learning and improvement subsequent to each problem-solving instance.
The objective was to develop an E-Learning platform that facilitates collaboration between students and teachers in the context of courses and classes. The platform utilizes a crowdsourcing approach for both the generation of teacher questions and the assignment of points among students. Moreover, students are encouraged to persist in problem-solving through continuous feedback.
Reviewing different workshops and conferences
EMNLP ACL SpLU-RoboNLP (ACL workshop)Having acquired foundational knowledge through courses and seminars, it is imperative to gain practical experience to excel in any field. To this end, we have enhanced our interns' proficiency through hands-on internship projects, providing exposure to real-world challenges and applications.
Android developer Web developer Digital marketing UI/UX designer Python learning systemThe seminars were organized as part of our CS-Plus Organization. The primary objective was to gather insights from expert programmers in various programming languages, with the intention of providing valuable tips and guidance to freshmen students.
Every programming languages involved in this courses and classes was primarily discussed in syntax and then applications. we aimed to help students learn how to develop their ideas with the presented tools.
PHP Android Java Python Matlab Ruby Django LaravelTo progress in your learning journey, it's crucial to have a well-defined understanding of the path you've embarked upon. These seminars aimed to introduce each significant job title and provide guidance on how to adequately prepare for those roles.
Game developer Android developer Web developer Digital marketing GraphistWe arranged this camp in order to make students familiar with the concept and life style of startup.