Welcome to my website! My name is Elaheh Raisi and I am currently machine learning engineer working at LinkedIn. Befor that I was postdoctoral researcher at Michigan State University working with professor Parisa Kordjamshidi. I was also a postdoctoral researcher at Brown University working with professor Stephen Bach. I received my Ph.D. degree in May 2019 under the supervision of professor Bert Huang.
My research interests span a wide variety of subfields of machine learning: learning with weak supervision, structured prediction, deep learning, NLP, computer vision, computational social science, Fairness and Explainability in ML, etc.
During my postdoc at MSU, I was working on integration of domain knowledge into statistical learning on complicated tasks with image and language involved; specifically, referring expression recognition.
During my postdoc at Brown, I was working on designing algorithms to learn when we do not have enough labeled data or information about the label of data. For problems such as fine-grained visual categorization that suffer from lack of labeled data, we introduced a multi-task learning based model to jointly train the target and other related task simultaneously. To construct the auxiliary data, we leverage a knowledge graph to query for semantically related concepts that are grounded in labeled image. We are extending this work to include more related auxiliary data to the target task and select more related data.
During my Ph.D., my primary research objective was to address the computational challenges associated with designing automated machine learning approaches for harassment-based cyberbullying detection. We have developed a weakly supervised framework, co-trained ensemble, in which two learning algorithms co-train one another, seeking consensus on whether examples in unlabeled data are cases of cyberbullying or not. One learner looks at language content in the message; another learner considers social structure to discover bullying. When designing our general framework, we address three tasks: First, using minimal supervision to learn the complex patterns of cyberbullying. Second, incorporating the efficacy of distributed representations of words and nodes. Finally, decreasing the sensitivity of the framework to language describing particular social groups including race, gender, religion, and sexual orientations.
You can reach me by: raisiela[at]msu.edu, elaheh[at]vt.edu