Md. Samiul Alim — Publications

📚 All Publications

Peer-reviewed journal articles, conference papers, workshop papers, manuscripts under review, and work in preparation. Underlined bold denotes my name. All published work is also listed on Google Scholar.

Summary: 6 peer-reviewed publications · 3 Q1 journal articles (1 first author) · 3 first-author conference papers · 1 workshop paper (NeurIPS 2026 MusIML) · 3 manuscripts under review · 1 in preparation.

📘 Peer-Reviewed Journal Articles

[J1] Multi-Teacher Knowledge Distillation and Ensemble Algorithm for Efficient Brain Tumor Classification in Resource-Constrained Environments With Explainable AI
Md. Samiul Alim, M. S. Tamim, S. Sarkar, M. A. Yousuf, A. S. Al-Moisheer, S. A. Alyami, M. A. Moni
Distils several teacher networks into a compact ensemble for brain-tumor classification on resource-constrained hardware, with explainable-AI overlays.
Knowledge-Based Systems, 2025 (Q1, IF 8.0) ▪ [DOI] ▪ Medical AI ▪ Knowledge Distillation
[J2] CAT-GS: Calibrated Adaptive Gating for Balanced and Robust Multimodal Learning
M. S. Tamim, S. Khan, Md. Samiul Alim, T. A. Khan, S. Rahman, N. Mohammed
A calibrated gating mechanism that balances modality contributions so that multimodal models stay robust when one modality dominates or degrades.
Neurocomputing, 2026 (Q1, IF 6.7) ▪ [arXiv] ▪ Multimodal Learning
[J3] Ensemble and Temporal Feature-Based Framework for Rainfall Classification in Bangladesh
M. S. Tamim, Md. Samiul Alim, T. A. Khan, M. Rahman, M. M. Anwar
Combines ensemble learning with temporal feature engineering to classify rainfall regimes from meteorological records in Bangladesh.
PLOS ONE, 2026 (Q1, IF 2.6) ▪ [Article] ▪ Applied ML

📰 Peer-Reviewed Conference Papers

[C1] Can LLMs Follow the Pulse of a Crisis? Evaluating Crisis Sentiment in Bangladesh’s July Uprising
Md. Samiul Alim, M. S. Tamim, T. A. Khan, S. Khan, R. F. Duti, S. Z. Ridoy, M. A. Moni
Evaluates whether large language models can track public sentiment on social media through the course of a national crisis.
AACL 2026, Proc. Asia-Pacific Chapter of the Association for Computational Linguistics, Hengqin, China ▪ [arXiv] ▪ Accepted ▪ NLP ▪ LLM Evaluation
[C2] Teacher-Guided One-Shot Pruning via Context-Aware Knowledge Distillation
Md. Samiul Alim, S. Khan, A. Biswas, F. Rahman, S. Rahman, N. Mohammed
Uses context-aware knowledge distillation from a teacher to prune a network in a single shot, avoiding iterative prune-and-retrain cycles.
IEEE BigData 2025, Proc. IEEE International Conference on Big Data, Macau, China (CORE B, 18.8% acceptance) ▪ [IEEE Xplore] ▪ Model Compression
[C3] When a Nation Speaks: Machine Learning and NLP in People’s Sentiment Analysis During Bangladesh’s 2024 Mass Uprising
Md. Samiul Alim, S. T. Mahir, R. Maisha, A. K. Tanvir, A. Md Mushfique
Curates a large multi-platform corpus from the 2024 mass uprising and benchmarks classical and transformer-based models on context-shifted political sentiment.
IEEE ICCIT 2025, Proc. IEEE International Conference on Computer and Information Technology ▪ [arXiv] ▪ NLP ▪ Social Science

🧪 Peer-Reviewed Workshop Papers

[W1] Compliance–Correctness Decomposition for Post-Training Evaluation of Small Language Models
K. Tshering, S. Das, S. Z. Ridoy, Md. Samiul Alim, A. T. Wasi, D.-K. Chae, M. A. Moni
Separates instruction compliance from answer correctness when evaluating post-trained small language models.
NeurIPS 2026 MusIML Workshop ▪ Accepted ▪ Small Language Models ▪ LLM Evaluation

⏳ Manuscripts Under Review

[U1] Retrieval Geometry Shapes Cache-Based CLIP Adaptation
Diagnoses the geometry of the memory space used by cache-based test-time adaptation of CLIP and how it governs reliable retrieval.
ICLR 2027 (CORE A*), under review ▪ In collaboration with Prof. Alex Lamb, College of AI, Tsinghua University ▪ Test-Time Adaptation
[U2] SphereAD: Hyperspherical Self-Reference for Training-Free Anomaly Localization Across Diverse Domains
Localizes anomalies without training by comparing features against a hyperspherical self-reference, transferring across visual domains.
WACV 2026 (CORE A), IEEE/CVF Winter Conference on Applications of Computer Vision, under review ▪ Computer Vision
[U3] When Bytes Take It All: CLIPByte for Tokenizer-Free Captioning via Byte-Level Vision–Language Bridging
Bridges a CLIP ViT-B/32 encoder with a ByT5-small byte-level decoder for open-vocabulary captioning in six typologically diverse languages, removing tokenizer-induced out-of-vocabulary bottlenecks.
WACV 2026 (CORE A), IEEE/CVF Winter Conference on Applications of Computer Vision, under review ▪ Vision–Language ▪ Multilingual

✍️ Manuscripts in Preparation

[P1] Constructive KAN: Growing Kolmogorov–Arnold Networks for Symbolic Regression
Neuroevolutionary optimization strategies for adaptive architecture and parameter learning in Kolmogorov–Arnold Networks (KANs).
In preparation ▪ Symbolic Regression