One paper has been accepted in ECAI 2024

/in /by

Congratulations to our senior project student Fahim Ahmed and research assistant Md Fahim for getting their paper accepted into the core rank A conference, European Conference on AI (ECAI) https://www.ecai2024.eu/ . The acceptance rate was very competitive (24%) this time for ECAI 2024. The title of the paper is, “Improving the Performance of Transformer-based Models Over Classical Baselines in Multiple Transliterated Languages”.

Here is a short description of the paper:

Online discourse, by its very nature, is rife with transliterated text along with code-mixing and code-switching. Transliteration is heavily featured due to the ease of inputting romanized text with standard keyboards over native scripts. Due to its ubiquity, it is a critical area of study to ensure NLP models perform well in real-world scenarios.

In this paper, we analyze the performance of various language model’s performance on classification of romanized/transliterated social media text. We chose the tasks of sentiment analysis and offensive language identification. We carried out experiments for three different languages, namely Bangla, Hindi, and Arabic (for six datasets). To our surprise, we discovered across multiple datasets that the classical machine learning methods (Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and XGBoost) perform very competitively with fine-tuned transformer-based mono / multilingual language models (BanglishBERT, HingBERT, and DarijaBERT, XLM-RoBERTa, mBERT, and mDeBERTa), tiny LLMs (Gemma-2B, and TinyLLaMa) and ChatGPT for classification tasks in transliterated text. Additionally, we investigated various mitigation strategies such as translation and augmentation via the use of ChatGPT, as well as Masked Language Modelling to dataset-specific pretraining for language models. Depending on the dataset and language, employing those mitigation techniques yields a 2-3% further improvement in accuracy and macro-F1 above baseline.

We demonstrate TF-IDF and BoW-based classifiers achieve performance within around 3% of fine-tuned LMs and thus could thus be considered as a strong baseline for transliterated text-based NLP tasks.

5 papers from CCDS has been accepted in ICPR 2024

/in /by

1.

Dehan, Farhan Noor; Fahim, Md; Rahman, AKM Mahabubur; Amin, M Ashraful; Ali, Amin Ahsan

TinyLLM Efficacy in Low-Resource Language 

In: 27th International Conference on Pattern Recognition, ICPR IEEE, KolKata, India, 2024.

2.

Sultana, Faria; Fuad, Md Tahmid Hasan; Fahim, Md; Rahman, Rahat Rizvi; Hossain, Meheraj; Amin, M Ashraful; Rahman, AKM Mahabubur; Ali, Amin Ahsan

How Good are LM and LLMs in Bangla Newspaper Article Summarization? 

In: 27th International Conference on Pattern Recognition, ICPR IEEE, KolKata, India, 2024.

3.

Kim, Minha; Bhaumik, Kishor; Ali, Amin Ahsan; Woo, Simon

MIXAD: Memory-Induced Explainable Time Series Anomaly Detection 

In: 27th International Conference on Pattern Recognition, ICPR IEEE, KolKata, India, 2024.

4.

Bhaumik, Kishor; Kimb, Minha; Niloy, Fahim Faisal; Ali, Amin Ahsan; Woo, Simon

SSMT: Few-Shot Traffic Forecasting with Single Source Meta-Transfer Learning 

In: IEEE Int’l Conf on Image Processing, ICPR IEEE, Abu Dhabi, 2024.

5.

Hossain, Mir Sazzat; Rahman, AKM Mahbubur; Amin, Md. Ashraful; Ali, Amin Ahsan

Lightweight Recurrent Neural Network for Image Super-resolution 

In: IEEE Int’l Conf on Image Processing, IEEE IEEE, Abu Dhabi, 2024.

Accepted Paper: Radio Galaxy Classification at INNS DLIA 2023

/in , /by

Our research paper, ‘Morphological Classification of Radio Galaxies using Semi-Supervised Group Equivariant CNNs,’ has been accepted for presentation at the esteemed INNS Deep Learning Innovations and Applications (INNS DLIA 2023) workshop, which is part of the International Joint Conference on Neural Networks (IJCNN 2023). The paper will also be published in the renowned Procedia Computer Science journal!

In this study, we tackled the challenge of limited labeled data in radio galaxy classification by employing a cutting-edge semi-supervised learning approach. By harnessing the power of Group Equivariant Convolutional Neural Networks (G-CNNs) as encoders, we achieved impressive results in classifying radio galaxies into the well-known Fanaroff-Riley Type I (FRI) and Type II (FRII) categories. [Link to Paper]

Explainable Hate Speech Detection: ICML 2023 Workshop Acceptance

/in , /by

Recently, Md Fahim, RA of CCDS with co-authors from UToronto, IUT and Fordham University has a paper accepted in AI and HCI workshop of ICML 2023. The paper proposes an interpretability and explainability oriented model to detect hate speech utilizing the pre-trained large language models. It creates dynamic class specific conceptual subspaces from which class specific attention is obtained by projecting the contextual embedding onto those spaces. These attentions provide better explainability of the detection task.
Paper Link: HateXplain2.0: An Explainable Hate Speech Detection Framework Utilizing Subjective Projection from Contextual Knowledge Space to Disjoint Concept Space

New paper from Dr. Ghosh’s group

/in , /by

Title: Holographic QFTs on AdS$_d$, wormholes and holographic interfaces.

arXiv preprint: https://arxiv.org/abs/2209.12094

We consider three related topics: (a) Holographic quantum field theories on AdS spaces. (b) Holographic interfaces of flat space QFTs. (c) Wormholes connecting generically different QFTs. We investigate in a concrete example how the related classical solutions explore the space of QFTs and we construct the general solutions that interpolate between the same or different CFTs with arbitrary couplings. The solution space contains many exotic RG flow solutions that realize unusual asymptotics, as boundaries of different regions in the space of solutions. We find phenomena like “walking” flows and the generation of extra boundaries via “flow fragmentation”.