We present a general framework for designing efficient algorithms for unsupervised learning problems, such as mixtures of Gaussians and subspace clustering. Our framework is based on a meta algorithm that learns arithmetic circuits in the presence of noise, using lower bounds. This builds upon the recent work of Garg, Kayal and Saha (FOCS 20), who designed such a framework for learning arithmetic circuits without any noise. A key ingredient of our meta algorithm is an efficient algorithm for a novel problem called Robust Vector Space Decomposition. We show that our meta algorithm works well when certain matrices have sufficiently large smallest non-zero singular values. We conjecture that this condition holds for smoothed instances of our problems, and thus our framework would yield efficient algorithms for these problems in the smoothed setting.

Jakub Macina, Nico Daheim, Lingzhi Wang, Tanmay Sinha, Manu Kapur, Iryna Gurevych, Mrinmaya Sachan. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics. 2023.

Socratic questioning is an educational method that allows students to discover answers to complex problems by asking them a series of thoughtful questions.Generation of didactically sound questions is challenging, requiring understanding of the reasoning process involved in the problem.We hypothesize that such questioning strategy can not only enhance the human performance, but also assist the math word problem (MWP) solvers.In this work, we explore the ability of large language models (LMs) in generating sequential questions for guiding math word problem-solving.We propose various guided question generation schemes based on input conditioning and reinforcement learning.On both automatic and human quality evaluations, we find that LMs constrained with desirable question properties generate superior questions and improve the overall performance of a math word problem solver.We conduct a preliminary user study to examine the potential value of such question generation models in the education domain.Results suggest that the difficulty level of problems plays an important role in determining whether questioning improves or hinders human performance.We discuss the future of using such questioning strategies in education.https://github.com/eth-nlped/ scaffolding-generation

In this work, we focus on automatically recognizing social conversational strategies that in human conversation contribute to building, maintaining or sometimes destroying a budding relationship.These conversational strategies include self-disclosure, reference to shared experience, praise and violation of social norms.By including rich contextual features drawn from verbal, visual and vocal modalities of the speaker and interlocutor in the current and previous turn, we can successfully recognize these dialog phenomena with an accuracy of over 80% and kappa ranging from 60-80%.Our findings have been successfully integrated into an end-to-end socially aware dialog system, with implications for virtual agents that can use rapport between user and system to improve task-oriented assistance.

In this work, we track the interaction of students across multiple Massive Open Online Courses (MOOCs) on edX. Leveraging the ``burstiness" factor of three of the most commonly exhibited interaction forms made possible by online learning (i.e, video lecture viewing, coursework access and discussion forum posting), we take on the task of predicting student performance (operationalized as grade) across these courses. Specifically, we utilize the probabilistic framework of Conditional Random Fields (CRF) to formalize the problem of predicting the sequence of grades achieved by a student in different MOOCs, taking into account the contextual dependency of this outcome measure on students' general interaction trend across courses. Based on a comparative analysis of the combination of interaction features, our best CRF model can achieve a precision of 0.581, recall of 0.660 and a weighted F-score of 0.560, outweighing several baseline discriminative classifiers applied at each sequence position. These findings have implications for initiating early instructor intervention, so as to engage students along less active interaction dimensions that could be associated with low grades.

With the proliferation of social media into our daily lives, online communities have become an important platform for collaborative learning and education. To connect users with varying knowledge levels and increase the net learning throughput, these communities often follow a question-answer based approach. Understanding what drives attention to help-seeking questions can reduce the amount of questions that go unnoticed or remain unanswered by the community. In this paper we discuss an important feature that affects the activity of the community, namely the community norms. We present a machine learning based trigger-driven feedback model that functions by (i) differentiating between help-seeking questions and follow-up posts – i.e. posts that are part of an ongoing discussion, and (ii) a dynamic intervention scheme to help improve question formulation. Our findings show that adhering to the community norms significantly increases the chance of eliciting a response.