Papers by Moin Nabi
EaSe: A Diagnostic Tool for VQA based on Answer Diversity (2021.naacl-main)
Copied to clipboard
| Challenge: | EASE is a diagnostic tool for Visual Question Answering (VQA) it quantifies the difficulty of an image, question sample. |
| Approach: | They propose a diagnostic tool which quantifies the difficulty of an image, question sample. |
| Outcome: | The proposed tool can be used to select the most-informative samples for training/fine-tuning. |
Contrastive Self-Supervised Learning for Commonsense Reasoning (2020.acl-main)
Copied to clipboard
| Challenge: | Existing methods for commonsense reasoning are limited by current methods . empirical results show that our method alleviates the limitation of current supervised approaches . |
| Approach: | They propose a self-supervised method to solve pronoun disambiguation problems . they leverage a mutual exclusive loss regularized by a contrastive margin to achieve commonsense reasoning . |
| Outcome: | The proposed method performs well on many NLP benchmarks. |
Attention-based Contrastive Learning for Winograd Schemas (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to learn discriminative features using contrastive objective are lacking. |
| Approach: | They propose a self-supervised framework that leverages a contrastive loss directly at the level of self-attention. |
| Outcome: | The proposed framework outperforms all comparable unsupervised approaches while occasionally surpassing supervised ones. |
SCD: Self-Contrastive Decorrelation of Sentence Embeddings (2022.acl-short)
Copied to clipboard
| Challenge: | Existing methods for self-supervised learning of representations are based on contrastive learning. |
| Approach: | They propose a self-supervised approach that optimizes a joint decorrelation and self-contrastive objective by leveraging the contrast arising from standard dropout at different rates. |
| Outcome: | The proposed method achieves comparable results with state-of-the-art methods on multiple benchmarks without using contrastive pairs. |
miCSE: Mutual Information Contrastive Learning for Low-shot Sentence Embeddings (2023.acl-long)
Copied to clipboard
| Challenge: | Existing methods for few-shot sentence embeddings are not robust enough to measure sentence similarity due to the ambiguity and variability of linguistic expressions. |
| Approach: | They propose a mutual information-based contrastive learning framework that imposes alignment between different views during contrastive training. |
| Outcome: | The proposed framework shows strong performance in few-shot learning domain compared to state-of-the-art methods, but comparable in full-shot scenario. |
Contrastive Perplexity for Controlled Generation: An Application in Detoxifying Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to generate toxic content by large language models are based on pipelines . current approaches focus on preserving performance while effectively mitigating toxicity . |
| Approach: | They propose a framework for implicit knowledge editing and controlled text generation by using hard negatives. |
| Outcome: | The proposed framework significantly reduces toxic generation while maintaining strong performance on downstream tasks. |
Attention Is (not) All You Need for Commonsense Reasoning (P19-1)
Copied to clipboard
| Challenge: | Recent language models such as word2vec have produced impressive results on various tasks such as question-answering and natural language inference. |
| Approach: | They propose a simple re-implementation of BERT for commonsense reasoning . they propose to use attention-guided reasoning to solve the Pronoun Disambiguation Problem . |
| Outcome: | The proposed model outperforms the state-of-the-art on several language understanding benchmarks while outperforming the existing models by a margin. |
Towards Zero-shot Commonsense Reasoning with Self-supervised Refinement of Language Models (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing language models can be refined for zero-shot commonsense reasoning . however, commons sense reasoning is still an unsolved problem . |
| Approach: | They propose a self-supervised learning approach that refines a pre-trained language model to boost conceptualization. |
| Outcome: | The proposed approach boosts conceptualization by utilizing loss landscape refinement. |