scription
← ideation

Thesis Title Ideation — July 7, 2026

Generated from 1,000 real papers (20 arXiv queries × 50 results). Each title is derived from a gap, limitation, or underexplored direction visible in the scraped literature.

🧠 NLP & Low-Resource Language Processing

  1. NusaTune: Parameter-Efficient Fine-Tuning of Indonesian LLMs via Adapter-Based Transfer Learning — Gap: NusaCrowd shows data scarcity; no PEFT study exists for Indonesian LLMs.
  2. IndoLoRA: Domain-Adaptive Low-Rank Adaptation for Under-Resourced Indonesian Language Models — Gap: LoRA-FAIR (arXiv 2411.14961) and FSLoRA (arXiv 2501.19389) focus on English; no Indonesian LoRA benchmarks exist.
  3. A Benchmark Suite for Indonesian Natural Language Understanding: Bridging the Gap Between NusaCrowd and Cross-Lingual Transfer — Gap: NusaCrowd (arXiv 2207.10524) aggregates datasets but lacks a standardized NLU benchmark.
  4. Cross-Lingual Transfer Learning for Indonesian NER: A Comparative Study of Multilingual BERT, XLM-R, and IndoBERT — Gap: BanglaCoNER (arXiv 2303.09306) shows multilingual NER gaps; no systematic comparison for Indonesian entities.
  5. IndoEmo: Fine-Grained Emotion Recognition in Indonesian Social Media Using Transformer Ensembles — Gap: PO-EMO (arXiv 2003.07723) covers aesthetic emotions in poetry; no equivalent for Indonesian social media.
  6. Low-Resource Indonesian Speaker Diarization: Synthetic Data Augmentation with Neural TTS — Gap: Domain adaptation of pyannote (arXiv 2601.03684) shows 29% DER; still far from usable for real Indonesian conversations.
  7. IndoClinicalNER: A Named Entity Recognition System for Indonesian Electronic Health Records — Gap: Clinical NER for Indonesian is absent; all existing work targets English (arXiv 1908.09138, 1908.10261).
  8. NusaBERT: A Large Pre-Trained Language Model for Indonesian and Its Regional Languages — Gap: Multilingual Maya/PALO models (arXiv 2605.17152) exist but no dedicated Indonesian BERT trained on diverse regional data.
  9. Evaluation of RAG-Based Clinical Decision Support Systems in Indonesian Healthcare: A Benchmark — Gap: AR-RAG (arXiv 2506.06962) and RAG healthcare papers exist, but none test Indonesian-language clinical data.
  10. Code-Mixed Indonesian-English NER: Challenges and Solutions for Social Media Text — Gap: CMNEROne (arXiv 2206.07318) tackles code-mixed NER but only for Hindi-English; Indonesian-English is unexplored.
  11. IndoEmoBank: Building a Large-Scale Emotion Lexicon for Indonesian Using Distant Supervision — Gap: Sentiment analysis survey (arXiv 2005.11882) lacks Indonesian-specific resources.
  12. Few-Shot Named Entity Recognition for Indonesian Clinical Texts Using Prompt-Based Learning — Gap: Query-based NER (arXiv 1908.09138) shows promise but untested on Indonesian clinical data.
  13. A Comparative Analysis of LoRA, DoRA, and AdaLoRA for Indonesian LLM Fine-Tuning — Gap: Multiple PEFT techniques exist but no systematic comparison on Indonesian tasks (arXiv 2501.19389, 2411.14961).
  14. IndoNLI: A Natural Language Inference Dataset for Indonesian — Gap: NusaCrowd contains no NLI dataset; cross-lingual NLI is an open problem.
  15. Transfer Learning from English to Indonesian for Aspect-Based Sentiment Analysis — Gap: Cross-lingual ABSA has no Indonesian benchmark despite commercial demand.
  16. Domain Adaptation of Pre-trained Language Models for Indonesian Legal Text Processing — Gap: Indonesian NLP survey (arXiv 2304.02746) identifies legal domain as underexplored.
  17. IndoPara: A Parallel Corpus for Indonesian-English Machine Translation in Specialized Domains — Gap: Low-resource Indonesian survey (arXiv 2207.10524) confirms parallel corpus scarcity.
  18. Active Learning for Low-Resource Indonesian NER with Limited Annotations — Gap: BanglaCoNER (arXiv 2303.09306) shows annotation is bottleneck; active learning unexplored for Indonesian.
  19. Multimodal Hate Speech Detection in Indonesian Social Media Using Vision-Language Models — Gap: Multimodal LLM survey (arXiv 2306.13549) and Maya/PALO (arXiv 2605.17152) exist but not fine-tuned for Indonesian hate speech.
  20. IndoMedicalQA: A Question Answering Dataset for Indonesian Clinical Text — Gap: Clinical QA exists for English; no Indonesian equivalent despite 270M speakers.
  21. Parameter-Efficient Fine-Tuning for Multilingual NER: A Study on Indonesian and Regional Languages — Gap: PEFT for NER (arXiv 1908.09138, 1908.10261) studied for English/Chinese; no low-resource multilingual study.
  22. IndoBART: A Denoising Autoencoder for Indonesian Text Generation Tasks — Gap: mBART exists but not fine-tuned specifically for Indonesian generation.
  23. Synthetic Data Generation for Indonesian NLP: Evaluating LLM-Based Approaches Against Human Annotations — Gap: Synthetic data used in diarization (arXiv 2601.03684); systematic evaluation for text tasks is missing.
  24. IndoWSD: Word Sense Disambiguation for Indonesian Using Contextual Embeddings — Gap: WSD benchmarks in Indonesian are absent despite rich polysemy in the language.
  25. Robustness Evaluation of Indonesian LLMs Under Adversarial Attacks — Gap: PPTC-R benchmark (arXiv 2403.03788) evaluates robustness in English; no Indonesian equivalent.
  26. Federated Fine-Tuning of Indonesian LLMs Across Heterogeneous Edge Devices — Gap: FSLoRA (arXiv 2501.19389) proposes federated sketch LoRA but evaluation is on English tasks only.
  27. IndoFactVerify: A Benchmark for Fact-Checking Indonesian News Using Retrieval-Augmented Generation — Gap: RAG for healthcare shows promise (arXiv seminars); factual verification in Indonesian is underexplored.
  28. Dialect-Aware Indonesian ASR: Adapting Whisper for Javanese, Sundanese, and Balinese Speech — Gap: Indonesian speech diarization (arXiv 2601.03684) shows adaptation works; ASR for regional dialects is next step.
  29. Zero-Shot Cross-Lingual Summarization for Indonesian News Using Multilingual LLMs — Gap: RL+LLM survey (arXiv 2509.16679) discusses cross-lingual tasks; no Indonesian summarization benchmark.
  30. IndoNLPProgress: A Living Benchmark for Tracking Indonesian NLP Model Improvements — Gap: NusaCrowd (arXiv 2207.10524) calls for reproducible progress tracking; no living benchmark exists.

🔒 Cybersecurity & Intrusion Detection

  1. Hybrid Deep Learning Framework for Real-Time Web Server Log Anomaly Detection Using BERT and Autoencoders — Gap: Hybrid DL work (arXiv 2212.00966) integrates K-means+GANomaly+CNN; BERT-based web-log specific approach missing.
  2. XploitIDS: Explainable AI for Network Intrusion Detection with SHAP and LIME on Transformer Models — Gap: XAI for IDS is underexplored; arXiv 2604.00187 focuses on accessibility, not cybersecurity XAI.
  3. Deep Embedded Clustering for Unsupervised Cyber Threat Intelligence: A Benchmark on CIC-IDS2018 — Gap: Deep embedded clustering (arXiv 2212.00966) uses K-means; modern DEC methods untested on IDS benchmarks.
  4. BERT-IDS: A Transformer-Based Intrusion Detection System for Imbalanced Network Traffic — Gap: Intrusion detection BERT (arXiv 2212.00966) uses CNN not BERT; transformer-based IDS remains underexplored for imbalance.
  5. Continuous Learning for Intrusion Detection: Adaptive Feature Ranking in Streaming Network Data — Gap: Online feature ranking (arXiv 1803.00530) uses SVM; no deep learning counterpart for streaming IDS.
  6. Multi-Stage Anomaly Detection for IoT Botnet Traffic Using Transformer Encoders — Gap: IoT intrusion detection not covered in scraped cyber papers despite being a major threat vector.
  7. Adversarial Robustness of Transformer-Based Intrusion Detection Systems — Gap: RL+LLM survey (arXiv 2509.16679) and XAI work (arXiv 2604.00187) discuss adversarial robustness; no specific IDS robustness study.
  8. Clustering-Driven Labeling for Semi-Supervised Log Anomaly Detection — Gap: Clustering cyber log analysis papers use unsupervised; semi-supervised with cluster-guided labeling is promising.
  9. Graph Neural Networks for Network Intrusion Detection: Modeling Traffic Flows as Temporal Graphs — Gap: Graph-based IDS is growing field but absent from scraped BERT-centric queries.
  10. Federated Intrusion Detection for Cross-Organizational Threat Sharing with Differential Privacy — Gap: Federated+transfer learning survey (arXiv 2207.02337) covers privacy; no specific federated IDS framework.
  11. Explainable Zero-Day Attack Detection Using Prototype-Based Neural Networks — Gap: XAI for IDS (arXiv 2604.00187) lacks focus on zero-day detection explanation.
  12. Lightweight Transformer for Edge-Based Intrusion Detection in Resource-Constrained IoT Devices — Gap: MS-YOLO (arXiv 2509.21696) shows edge deployment success in vision; analog for IDS missing.
  13. Temporal Attention Networks for Multi-Step Cyberattack Prediction from Log Sequences — Gap: Web log deep learning papers cover detection; few address prediction of multi-step attacks.
  14. Unsupervised Log Parsing and Anomaly Detection Using Sentence-BERT Embeddings — Gap: Clustering cyber log analysis (arXiv papers) relies on TF-IDF or word2vec; SBERT for logs is underexplored.
  15. Benchmarking Explainable AI Methods for Intrusion Detection: A Comparative Study — Gap: XAI intrusion detection (arXiv 2604.00187) mentions other XAI methods but no systematic benchmark exists.
  16. Cross-Dataset Generalization of Deep Learning IDS: A Multi-Benchmark Evaluation — Gap: Hybrid framework (arXiv 2212.00966) evaluates on NSL-KDD, CIC-IDS2018, TON_IoT; cross-dataset generalization study missing.
  17. Contrastive Learning for Network Anomaly Detection with Limited Labeled Data — Gap: Semi-distributed IDS (arXiv 1006.1956) and hybrid approach (arXiv 2212.00966) don’t explore contrastive learning.
  18. Domain Adaptation for Intrusion Detection: Transferring Knowledge Between Network Environments — Gap: Road context-aware IDS (arXiv 1908.00732) shows domain-specificity; transfer between network environments untested.
  19. Time-Aware BERT for Sequential Intrusion Detection in Enterprise Networks — Gap: BERT for intrusion detection (arXiv papers) processes flows independently; temporal BERT is novel.
  20. Visual Analytics for Explainable Intrusion Detection: A Dashboard Framework for Security Analysts — Gap: XAI for BLV users (arXiv 2604.00187) suggests modality gaps; no visual analytics IDS dashboard studied.
  21. Reinforcement Learning for Adaptive Intrusion Response in Software-Defined Networks — Gap: RL+LLM survey (arXiv 2509.16679) covers RL applications; RL for active intrusion response is absent from scraped papers.
  22. Self-Supervised Representation Learning for Encrypted Traffic Classification — Gap: Deep embedded clustering (arXiv papers) struggles with encrypted traffic; self-supervised pre-training is promising.
  23. Multimodal Intrusion Detection: Combining Network Flows, Logs, and System Calls with Transformer Fusion — Gap: Multimodal LLM survey (arXiv 2306.13549) discusses fusion; no multimodal IDS using transformers exists.
  24. Cost-Sensitive Learning for Intrusion Detection in Highly Imbalanced Network Traffic — Gap: Online feature ranking (arXiv 1803.00530) doesn’t address class imbalance beyond feature selection.
  25. Evaluating Large Language Models for Cyber Threat Intelligence Extraction from Unstructured Reports — Gap: LLM evaluation survey (arXiv 2403.03788) focuses on PPT tasks; no CTI extraction benchmark.
  26. Interpretable Deep Learning for DNS Tunnel Detection Using Attention Visualization — Gap: XAI for intrusion detection (arXiv 2604.00187, arXiv 2212.00966) doesn’t address specific attack types like DNS tunneling.
  27. Few-Shot Intrusion Detection for Novel Attack Families Using Metric Learning — Gap: Hybrid DL framework (arXiv 2212.00966) requires normal data; few-shot for novel attacks is underexplored.
  28. Ensemble of Transformer and GNN for Host-Level Intrusion Detection on Audit Logs — Gap: Web server log anomaly (arXiv papers) lacks GNN-based host-level detection.
  29. Privacy-Preserving Federated Intrusion Detection Using Secure Aggregation and Differential Privacy — Gap: Federated+transfer survey (arXiv 2207.02337) covers defenses; no specific privacy-preserving IDS architecture.
  30. Generative Adversarial Networks for Data Augmentation in Intrusion Detection: A Systematic Evaluation — Gap: DCGAN medical image papers (arXiv 2412.01496) show GANs for augmentation; no such study for IDS.

👁️ Computer Vision & Medical Imaging

  1. MS-YOLO-Edge: Optimized Real-Time Object Detection for Low-Power Edge Devices — Gap: MS-YOLO (arXiv 2509.21696) achieves 6.7 GFLOPs; further edge optimization is open.
  2. Federated Object Detection for Privacy-Preserving Visual Surveillance Across Heterogeneous Cameras — Gap: Federated+transfer survey (arXiv 2207.02337) discusses vision but not YOLO-specific federated learning.
  3. Transfer Learning with YOLOv8 for Indonesian Traffic Sign Detection in Adverse Weather Conditions — Gap: MS-YOLO uses FLIR ADAS (infrared); RGB-based Indonesian traffic sign detection is untapped.
  4. Synthetic-to-Real Domain Adaptation for AI-Generated Image Detection — Gap: Synthetic AI image detection papers identify domain shift as key challenge.
  5. DCGAN-Based Data Augmentation for Class-Imbalanced Medical Image Classification — Gap: FRD (arXiv 2412.01496) evaluates generative models; augmentation for class imbalance is underexplored.
  6. IndoVLM: A Vision-Language Model for Indonesian Visual Question Answering — Gap: Maya/PALO (arXiv 2605.17152) target multilingual VLMs; no Indonesian-specific VLM exists.
  7. Multimodal Fake News Detection in Indonesian Social Media Using CLIP and BERT Fusion — Gap: Multimodal LLM survey (arXiv 2306.13549) covers fake news; no Indonesian context.
  8. Evaluating Fréchet Radiomic Distance for MRI Quality Assessment in Low-Resource Clinical Settings — Gap: FRD (arXiv 2412.01496) introduces radiomic metrics; deployment in low-resource settings is unexplored.
  9. Few-Shot Object Detection for Rare Indonesian Wildlife Species Using YOLO with Siamese Networks — Gap: Transfer learning YOLO papers don’t cover few-shot detection for rare species.
  10. Lightweight GAN for Medical Image Synthesis on Mobile Health Platforms — Gap: DCGAN medical image papers (arXiv 2412.01496, 2111.10480) are compute-heavy; mobile deployment is open.
  11. Vision Transformer vs. YOLO: A Comparative Study for Real-Time Industrial Defect Detection — Gap: Oriented object detection survey (arXiv 2302.10473) compares methods; ViT vs YOLO on industrial data is missing.
  12. Cross-Modal Retrieval for Indonesian Radiology Reports Using Vision-Language Pretraining — Gap: Multimodal LLM tutorial (arXiv 2605.17152) covers low-resource VLMs; radiology + Indonesian is unexplored.
  13. Domain Generalization for Medical Image Segmentation: A Benchmark Across Unseen Hospital Protocols — Gap: FRD (arXiv 2412.01496) studies domain gaps; generalization across hospitals is open.
  14. YOLO-Based Real-Time Pothole Detection for Indonesian Road Infrastructure Monitoring — Gap: Indonesian-specific object detection application with no existing benchmark.
  15. Explaining Medical Image Classifications Using Concept Attribution on GAN-Generated Counterfactuals — Gap: DCGAN medical (arXiv papers) + XAI (arXiv 2604.00187); no combined counterfactual explanation for medical images.
  16. Multilingual Image Captioning for Indonesian Using Vision-Language Models with Adapter Tuning — Gap: Multilingual MM-LLMs (arXiv 2605.17152) lack Indonesian image captioning benchmark.
  17. Self-Supervised Pretraining for Medical Image Analysis: A Comparison of Contrastive Methods on Chest X-Rays — Gap: TransMorph (arXiv 2111.10480) uses self-supervised registration; broader pretraining for classification is open.
  18. Adversarial Robustness of YOLO Object Detectors in Autonomous Driving Scenarios — Gap: MS-YOLO (arXiv 2509.21696) optimizes for edge; adversarial robustness in autonomous driving not addressed.
  19. Synthetic Medical Image Generation with StyleGAN for Privacy-Preserving Data Sharing — Gap: DCGAN medical (arXiv 2412.01496) focuses on metrics; StyleGAN’s utility for medical data sharing is a gap.
  20. Real-Time Multi-Object Tracking Across Camera Networks Using YOLO+DeepSORT — Gap: YOLO transfer learning papers focus on detection; multi-camera tracking is an extension gap.
  21. Annotation-Efficient Medical Image Segmentation Using Box-Supervised Learning and SAM — Gap: TransMorph (arXiv 2111.10480) requires full annotations; box-supervised segmentation not covered.
  22. Indonesian Cultural Heritage Object Detection: A Dataset and Benchmark Using YOLOv10 — Gap: No Indonesian cultural heritage computer vision dataset exists in the literature.
  23. DCGAN for Retinal Image Enhancement: Improving Diagnostic Quality in Low-Resource Eye Clinics — Gap: DCGAN medical (arXiv papers) covers general medical images; retinal-specific application is missing.
  24. Edge-Cloud Collaborative Object Detection for Smart City Surveillance in Developing Countries — Gap: MS-YOLO (arXiv 2509.21696) is edge-only; edge-cloud collaborative splitting is a gap.
  25. Vision-Language Model Fine-Tuning for Agricultural Disease Detection in Indonesian Rice Fields — Gap: Multimodal tutorial (arXiv 2605.17152) discusses low-resource VLMs; agriculture is untapped.
  26. Uncertainty-Aware YOLO for Reliable Object Detection in Low-Visibility Conditions — Gap: MS-YOLO (arXiv 2509.21696) doesn’t quantify prediction uncertainty.
  27. Foundation Model for Indonesian Medical Imaging: Adapting SAM for Local Anatomical Variations — Gap: SAM adapters exist for general medical; Indonesian-specific anatomy not covered.
  28. Continual Learning for Object Detection: Avoiding Catastrophic Forgetting in YOLO Models — Gap: Federated+transfer survey (arXiv 2207.02337) covers continual learning; object detection specific study is missing.
  29. Evaluating Perceptual Metrics for Medical Image Generation: FID vs. FRD Across Modalities — Gap: FRD (arXiv 2412.01496) proposes new metric; comprehensive comparison across modalities is a gap.
  30. Weakly Supervised Object Detection for Indonesian Traffic Surveillance Using Image-Level Labels — Gap: YOLO transfer papers require bounding box annotations; weakly supervised approach reduces annotation cost.

📊 LLM Evaluation & Benchmarking

  1. IndoEval: A Comprehensive Evaluation Suite for Indonesian Large Language Models — Gap: LLM eval survey (arXiv 2403.03788, 2509.16679) covers English; no Indonesian LLM evaluation benchmark exists.
  2. RobustnessBench: Evaluating LLM Task Completion Robustness Under Real-World Input Perturbations — Gap: PPTC-R (arXiv 2403.03788) evaluates PPT tasks; general task robustness benchmark is open.
  3. Reinforcement Learning from Human Feedback for Indonesian LLM Alignment — Gap: RL+LLM survey (arXiv 2509.16679) discusses RLHF; no Indonesian RLHF dataset or study exists.
  4. Benchmarking LLM Safety and Alignment in Bahasa Indonesia: A Red-Teaming Approach — Gap: LLM alignment safety papers don’t cover Indonesian cultural context.
  5. Cross-Lingual Transfer Evaluation of Reasoning Capabilities: From English LLMs to Indonesian — Gap: LLM evaluation benchmarks (arXiv 2509.16679) don’t test cross-lingual reasoning degradation.
  6. IndoMT-Bench: A Multi-Task Benchmark for Indonesian Language Model Evaluation — Gap: Inspired by MT-Bench; no Indonesian multi-turn evaluation suite.
  7. Evaluating Hallucination Rates in Indonesian LLMs for Healthcare Applications — Gap: Healthcare RAG papers don’t evaluate hallucination in Indonesian context.
  8. Automated Benchmark Generation for Indonesian NLP Tasks Using LLM Distant Supervision — Gap: NusaCrowd (arXiv 2207.10524) calls for more datasets; LLM-generated benchmarks are a potential solution.
  9. Dynamic Benchmarking: Continual Evaluation of LLMs on Evolving Indonesian News and Social Media — Gap: PPTC-R (arXiv 2403.03788) uses static benchmarks; dynamic benchmarks keep pace with data drift.
  10. IndoInstruct: A Large-Scale Instruction Tuning Dataset for Indonesian LLMs — Gap: Instruction tuning datasets exist for English/Chinese; no equivalent for Indonesian.
  11. Evaluating Cultural Bias in Multilingual LLMs: A Case Study on Indonesian Social Norms — Gap: RL+LLM survey (arXiv 2509.16679) discusses alignment; cultural bias evaluation is a gap.
  12. Cost-Effective Evaluation: Predicting LLM Performance on Indonesian Tasks Using Proxy Metrics — Gap: LLM evaluation is expensive; proxy metrics for low-resource languages not explored.
  13. Long-Context LLM Evaluation for Indonesian Legal Document Comprehension — Gap: LLM eval survey covers general tasks; long-context understanding in Indonesian documents is untested.
  14. Adversarial Evaluation of LLM Safety Guardrails in Indonesian Context — Gap: LLM alignment safety papers test English adversarial prompts; Indonesian-specific guardrails underexplored.
  15. Human Evaluation Framework for Indonesian LLM Outputs: Fluency, Factuality, and Cultural Appropriateness — Gap: PPTC-R (arXiv 2403.03788) uses automated; human evaluation framework for Indonesian is missing.
  16. IndoRAG: A Retrieval-Augmented Generation Benchmark for Indonesian Question Answering — Gap: AR-RAG (arXiv 2506.06962) and RAG healthcare papers don’t test Indonesian data.
  17. Sample Efficiency of RLHF for Low-Resource Language Model Alignment — Gap: RL+LLM survey (arXiv 2509.16679) doesn’t address sample efficiency in low-resource settings.
  18. Benchmarking Tool-Augmented LLMs for Indonesian Real-World Tasks — Gap: PPTC-R (arXiv 2403.03788) uses PPT tools; general tool use evaluation in Indonesian is missing.
  19. IndoMath: A Mathematical Reasoning Benchmark for Indonesian LLMs — Gap: No Indonesian math reasoning dataset; cross-lingual transfer of reasoning is a known gap.
  20. Evaluating Multimodal LLMs on Indonesian Cultural Visual Question Answering — Gap: Multimodal LLM survey (arXiv 2306.13549) lacks culture-specific VQA benchmarks.

🔗 Cross-Domain & Survey-Based

  1. A Survey of Indonesian Natural Language Processing in the LLM Era: Progress, Challenges, and Opportunities — Gap: Survey (arXiv 2304.02746) covers pre-LLM era; GPT-era survey needed.
  2. Deep Learning for Intrusion Detection in Indonesian Critical Infrastructure: A Systematic Review — Gap: Survey intrusion 2024 papers don’t focus on Indonesian infrastructure context.
  3. Computer Vision for Tropical Agriculture: A Survey of Deep Learning Applications in Indonesia — Gap: CV survey 2024 papers omit agricultural applications in tropical developing nations.
  4. Parameter-Efficient Fine-Tuning of LLMs: A Comprehensive Survey with a Focus on Low-Resource Languages — Gap: Existing PEFT surveys center on English; low-resource perspective is missing.
  5. Retrieval-Augmented Generation in Clinical Decision Support: A Systematic Review and Research Agenda — Gap: RAG healthcare papers lack systematic review of clinical deployment challenges.
  6. Explainable AI for Cybersecurity: From Post-Hoc Explanations to Inherently Interpretable Models — Gap: XAI IDS papers survey (arXiv 2604.00187, 2212.00966) but not from cybersecurity perspective.
  7. Transfer Learning and Domain Adaptation for Deep Learning in Southeast Asian Languages: A Survey — Gap: Federated+transfer survey (arXiv 2207.02337) is general; SEA language-specific survey needed.
  8. Generative AI for Medical Imaging in Low-Resource Settings: A Systematic Review — Gap: DCGAN medical (arXiv 2412.01496) and TransMorph (arXiv 2111.10480) target high-resource settings.
  9. Vision-Language Models for Document Understanding in Indonesian: A Survey and Benchmark — Gap: Multimodal LLM tutorial (arXiv 2605.17152); document understanding in Indonesian is underexplored.
  10. Federated Learning for NLP in Low-Resource Languages: Challenges and Opportunities — Gap: Federated+transfer survey (arXiv 2207.02337) covers security; low-resource NLP-specific review needed.
  11. From BERT to GPT: The Evolution of Indonesian Language Models — A Survey — Gap: Indonesian NLP survey (arXiv 2304.02746) published 2023; needs update to cover LLM era.
  12. Machine Unlearning in LLMs: A Survey of Methods and Evaluation Benchmarks for Safety and Privacy — Gap: LLM alignment safety papers discuss alignment but not targeted unlearning.
  13. Anomaly Detection in Time-Series Data: A Survey of Deep Learning Methods with Applications to Indonesian Systems — Gap: Web server log anomaly papers (arXiv 2212.00966) lack regional context application.
  14. Audio Deepfake Detection in Low-Resource Languages: A Survey — Gap: Synthetic AI image detection papers cover vision; audio deepfake detection for Indonesian is missing.
  15. Multimodal AI for Disaster Response in the Indonesian Archipelago: A Survey — Gap: Multimodal LLM survey (arXiv 2306.13549) doesn’t cover disaster response.
  16. LoRA Variants and Extensions: A Systematic Survey of Low-Rank Adaptation Methods — Gap: LoRA fine-tuning papers (arXiv 2501.19389, 2411.14961) cover specific methods; comprehensive survey missing.
  17. Evaluation Methodologies for Low-Resource NLP: A Critical Survey — Gap: LLM eval survey (arXiv 2509.16679) covers high-resource; methodology for low-resource eval is open.
  18. Deep Learning for Indonesian Sign Language Recognition: A Survey and Benchmark — Gap: CV survey 2024 papers omit sign language recognition for Indonesian.
  19. Cybersecurity in the Indonesian Digital Economy: A Survey of AI-Based Threat Detection Research — Gap: Intrusion detection survey 2024 covers global trends; Indonesia-specific survey needed.
  20. Synthetic Data for Low-Resource NLP: A Survey of Generation Methods and Quality Metrics — Gap: Synthetic AI detection papers focus on detection; generation methods for low-resource languages under-reviewed.

⚙️ Hybrid & Applied Systems

  1. IndoHealthChat: A Retrieval-Augmented LLM Chatbot for Indonesian Primary Healthcare — Gap: RAG healthcare papers don’t deploy in Indonesian primary care context.
  2. End-to-End Indonesian Hate Speech Detection with Multimodal Fusion of Text, Image, and Metadata — Gap: Multimodal LLM for hate speech; no existing Indonesian multimodal hate speech system.
  3. Real-Time Indonesian License Plate Recognition Using Lightweight YOLO and CRNN — Gap: YOLO transfer learning for ALPR; Indonesian plates have unique format.
  4. AI-Powered Indonesian Legal Document Reviewer Using Fine-Tuned LLMs with LoRA — Gap: LoRA fine-tuning (arXiv 2501.19389) in legal domain; no Indonesian legal AI exists.
  5. Privacy-Preserving Indonesian Sentiment Analysis Using Federated Learning with Differential Privacy — Gap: Federated+transfer survey (arXiv 2207.02337); no Indonesian sentiment federated study.
  6. Indonesian Fake News Detection System Using BERT-CNN Hybrid with Multi-Head Attention — Gap: BERT for intrusion detection (arXiv papers); fake news detection with similar architecture untested in Indonesian.
  7. Automated Essay Scoring for Indonesian National Exams Using Fine-Tuned Transformer Models — Gap: Indonesian NLP survey (arXiv 2304.02746) identifies education as priority; automated scoring gap exists.
  8. Smart Contract Vulnerability Detection in Indonesian Blockchain Applications Using Fine-Tuned LLMs — Gap: LoRA fine-tuning (arXiv 2411.14961) for code tasks; blockchain security in Indonesian context is novel.
  9. Indonesian Sign Language to Text Translation Using Vision Transformers and LLM Post-Processing — Gap: Vision transformer + LLM pipeline; no Indonesian sign language system exists.
  10. Adaptive Learning System for Indonesian Students Using LLM-Generated Personalized Content with PEFT — Gap: RL+LLM survey (arXiv 2509.16679); adaptive edtech in Indonesian education system not addressed.
  11. Indonesian Language Model for Anti-Money Laundering: Transaction Narrative Classification — Gap: Indonesian NLP for finance; no AML benchmark exists for Indonesian transaction texts.
  12. Automated Indonesian Medical Coding from Clinical Notes Using Fine-Tuned Clinical BERT — Gap: Clinical NER papers target English; ICD coding from Indonesian clinical text is a clear gap.
  13. Indonesian Social Media Crisis Monitor Using RAG-Augmented LLMs for Disaster Response — Gap: RAG healthcare (arXiv papers) in emergency response; Indonesia is disaster-prone, no system exists.
  14. Cross-Lingual Indonesian-Javanese-Sundanese Translation Using LoRA-Adapted Multilingual LLMs — Gap: Low-resource Indonesian translation (arXiv 2207.10524); regional language translation gap persists.
  15. AI-Powered Indonesian Agriculture Advisory System Using Vision-Language Models for Pest Detection — Gap: CV for agriculture; no Indonesian pest detection VLM exists.
  16. Indonesian Exam Question Generation Using LLMs with RAG from Curriculum Documents — Gap: RAG + generation (arXiv 2506.06962); educational content generation in Indonesian is an application gap.
  17. Context-Aware Indonesian Chatbot for Mental Health Support Using Fine-Tuned LLMs — Gap: Healthcare RAG (arXiv papers) for mental health; Indonesian mental health chatbot is absent.
  18. Semi-Automated Indonesian News Summarization with Human-in-the-Loop Quality Control — Gap: Cross-lingual summarization papers don’t cover Indonesian; HITL approach is practical gap.
  19. Indonesian Fisheries Monitoring: YOLO-Based Fish Species Detection for Sustainable Aquaculture — Gap: YOLO object detection for Indonesian fisheries; no dataset or benchmark exists.
  20. AI-Assisted Indonesian Manuscript Digitization: Handwriting Recognition Using Vision-Language Models — Gap: CV for historical document digitization; Indonesian scripts (Javanese, Balinese) underexplored.
  21. Indonesian Code-Switched Speech Recognition Using Self-Supervised Pretrained Models — Gap: Indonesian audio domain adaptation (arXiv 2601.03684) for diarization; code-switched ASR is next.
  22. Explainable Loan Default Prediction Using XAI-Augmented LLMs for Indonesian Banking — Gap: XAI (arXiv 2604.00187) for financial decision-making in Indonesian context is unexplored.
  23. Batik Motif Classification Using YOLO and Fine-Grained Visual Recognition — Gap: Cultural heritage CV; batik motif classification with deep learning lacks comprehensive benchmark.
  24. Indonesian Traffic Flow Prediction Using Hybrid CNN-Transformer with Graph Neural Networks — Gap: Deep learning for web logs (arXiv papers) doesn’t cover traffic prediction.
  25. Automated Paraphrase Generation for Indonesian Academic Writing Using Fine-Tuned LLMs — Gap: Indonesian NLP tasks (arXiv 2304.02746) don’t include paraphrase generation.
  26. Indonesian Food Recognition and Calorie Estimation Using Multimodal Vision-Language Models — Gap: No Indonesian food dataset for deep learning; multimodal estimation is novel.
  27. Collaborative Multi-Agent LLM System for Indonesian Healthcare Triage — Gap: RL+LLM agents (arXiv 2509.16679); multi-agent healthcare in Indonesian is novel.
  28. Indonesian E-Commerce Product Classification Using Multilingual BERT with Category Hierarchies — Gap: Indonesian NLP in e-commerce; product taxonomy classification is underexplored.
  29. Cross-Modal Indonesian Recipe Retrieval: From Image to Instruction Using Contrastive Learning — Gap: Multimodal retrieval (arXiv 2605.17152) for Indonesian food; no existing dataset.
  30. Indonesian Language Model for Contract Clause Extraction and Risk Analysis — Gap: NER for legal documents (arXiv 1908.09138); Indonesian contract analysis gap exists.
  31. Federated YOLO for Privacy-Preserving Surveillance in Indonesian Smart Cities — Gap: Federated+transfer (arXiv 2207.02337) + YOLO; privacy-preserving surveillance for Indonesian smart cities.
  32. Deep Learning for Indonesian Coral Reef Health Monitoring Using Underwater Image Analysis — Gap: Marine CV (arXiv 2311.14762 focus on maritime); coral reef health with deep learning in Indonesia.
  33. Indonesian Fake Voice Detection Using Spectral Features and Transformer Models — Gap: Synthetic detection (arXiv papers) for images/audio; Indonesian language voice deepfake detection.
  34. LLM-Based Indonesian Academic Paper Reviewer: Automated Feedback for Structure and Grammar — Gap: LLM evaluation tools for Indonesian academic writing quality.
  35. Indonesian Commodity Price Forecasting Using LLM-Augmented Time Series Models with News Sentiment — Gap: Multi-modal fusion (text + time series); Indonesian commodity markets.
  36. Domain-Specific Indonesian Chatbot for Islamic Religious Education Using RAG from Classical Texts — Gap: RAG (arXiv 2506.06962) for religious domain; Islamic texts in Indonesian.
  37. Automated Indonesian Tax Document Classification Using Fine-Tuned IndoBERT — Gap: Document understanding for Indonesian tax forms; government digitalization.
  38. Indonesian Traffic Violation Detection: A YOLO-Based System for Unmanned Traffic Enforcement — Gap: Transfer learning YOLO for traffic enforcement in Indonesian context.
  39. Cross-Lingual Embedding Alignment for Indonesian Regional Language Retrieval — Gap: Multilingual embedding alignment; no study for Indonesian regional languages.
  40. Indonesian Pronunciation Error Detection Using Fine-Tuned Wav2Vec 2.0 for Language Learning — Gap: Indonesian audio processing (arXiv 2601.03684) for language education.
  41. AI-Generated Indonesian News Detection: A Benchmark for Synthetic Text Identification — Gap: AI-generated text detection papers don’t test Indonesian generated content.
  42. Indonesian Land Cover Classification from Satellite Imagery Using Vision Transformers — Gap: Oriented object detection survey (arXiv 2302.10473) for remote sensing; Indonesian land cover dataset.
  43. Student Engagement Detection in Indonesian Online Classrooms Using YOLO-Based Facial Analysis — Gap: YOLO transfer learning (arXiv papers) for education analytics.
  44. Indonesian Parliamentary Speech Analysis: Topic Modeling and Stance Detection with Fine-Tuned LLMs — Gap: Indonesian NLP for political analysis; parliamentary speech dataset.
  45. Indonesian Legal Judgment Prediction Using Fine-Tuned LLMs with Structured Case Facts — Gap: Legal NLP for Indonesian court decisions; judgment prediction benchmark.
  46. Indonesian Folk Tale Generation Using LLMs with Cultural Knowledge Graph Integration — Gap: RL+LLM survey (arXiv 2509.16679) for creative generation; Indonesian folklore.
  47. Multimodal Emotion Recognition in Indonesian Video Content Using Audio-Visual-Text Fusion — Gap: Emotion recognition (arXiv 2003.07723) for Indonesian; no multimodal fusion study.
  48. Indonesian Waste Classification for Smart Recycling Using Lightweight CNN Models — Gap: CV for environmental applications in Indonesia; waste classification dataset.
  49. Active Learning for Indonesian NER in Specialized Domains with Minimal Expert Annotation — Gap: NER (arXiv 1908.09138) active learning for Indonesian; annotation bottleneck.
  50. Indonesian Traditional Dance Movement Recognition Using Pose Estimation and Sequence Models — Gap: CV survey 2024; cultural movement recognition in Indonesian dances.
  51. Indonesian News Article Attribution: Authorship Verification Using Stylometric Features with LLMs — Gap: AI-generated text detection (arXiv papers); authorship verification in Indonesian.
  52. Indonesian Hateful Meme Detection Using Multimodal Fusion of Vision-Language Models — Gap: Multimodal LLM (arXiv 2306.13549) for harmful content; Indonesian meme dataset.
  53. Grammatical Error Correction for Indonesian L2 Learners Using Fine-Tuned mT5 — Gap: Indonesian NLP (arXiv 2304.02746) for language learning; GEC dataset.
  54. Indonesian Lung Disease Classification from Chest X-Rays Using Vision Transformer Ensembles — Gap: FRD (arXiv 2412.01496) for chest X-ray; no Indonesian lung disease benchmark.
  55. Self-Supervised Indonesian Speech Representation Learning for Under-Resourced Regional Dialects — Gap: Wav2Vec self-supervised; regional Indonesian dialects (300+ languages) underexplored.
  56. Indonesian Argument Mining: Extracting Premises and Conclusions from Debate Transcripts — Gap: NLP for argumentation in Indonesian; existing resources cover English only.
  57. Privacy-Aware Indonesian Sentiment Analysis Using On-Device Fine-Tuned Small Language Models — Gap: LoRA fine-tuning (arXiv 2501.19389) + edge deployment; on-device sentiment.
  58. Indonesian Question Generation from Textbooks Using LLMs with Curriculum-Aware Prompting — Gap: RAG-based generation; Indonesian textbook question generation.
  59. Few-Shot Indonesian Intent Classification for Task-Oriented Dialogue Systems — Gap: Low-resource NLP (arXiv 2207.10524); intent classification for Indonesian dialogue.
  60. Indonesian Document Image Layout Analysis Using Graph Neural Networks — Gap: Document understanding; Indonesian document layout complexity.
  61. Indonesian Cyberbullying Detection on TikTok Comments Using Multimodal Analysis — Gap: Cyberbullying detection in Indonesian; multimodal (text + video metadata).
  62. Continual Pre-Training of Indonesian LLMs on Evolving News Corpora — Gap: Continual learning (arXiv 2207.02337); Indonesian news domain adaptation.
  63. Indonesian Radiology Report Generation from Chest X-Rays Using Vision-Language Pretraining — Gap: Medical VLM (arXiv papers); Indonesian radiology report generation.
  64. Deep Learning for Indonesian Peatland Fire Detection Using Satellite and Drone Imagery — Gap: Remote sensing CV; Indonesian environmental monitoring.
  65. Indonesian Poetry Generation Using Fine-Tuned LLMs with Meter and Rhyme Constraints — Gap: PO-EMO (arXiv 2003.07723) for poetry in Indonesian; generation with formal constraints.
  66. Cross-Lingual Entity Linking for Indonesian Texts: Connecting Local Entities to Wikipedia — Gap: Entity linking for Indonesian; no existing benchmark despite knowledge base gap.
  67. Indonesian Academic Text Paraphrase Detection Using Transformer Models — Gap: AI-generated text detection for Indonesian academic integrity.
  68. Indonesian Clinical Trial Matching: A RAG-Based System for Patient Recruitment — Gap: RAG healthcare (arXiv papers) for clinical trial matching in Indonesian hospitals.
  69. Indonesian Low-Resource Dependency Parsing Using Multilingual Transfer with Adapter Layers — Gap: PEFT for syntactic parsing; Indonesian dependency treebanks are scarce.
  70. Zero-Shot Indonesian Toxic Comment Classification Using Multilingual LLMs with Chain-of-Thought Prompting — Gap: Cross-lingual zero-shot for content moderation; Indonesian toxic language benchmark.
  71. Indonesian Social Media Bot Detection Using Graph Neural Networks and LLM-Enhanced Features — Gap: Cybersecurity for social media; Indonesian bot detection dataset.
  72. Deep Embedded Clustering for Indonesian News Topic Discovery in Real-Time Streams — Gap: Deep embedded clustering anomaly (arXiv papers) for topic detection; Indonesian news streams.
  73. Indonesian Financial Sentiment Analysis with Domain-Adapted LLMs Using LoRA — Gap: LoRA fine-tuning (arXiv 2411.14961) for finance; Indonesian financial sentiment.
  74. Automated Indonesian Essay Evaluation Using Multi-Task LLMs with Rubric-Aware Scoring — Gap: LLM evaluation (arXiv 2403.03788) for education; rubric-based scoring in Indonesian.
  75. Indonesian Language Model Watermarking: Protecting Copyright in Generated Text — Gap: AI-generated text detection (arXiv papers); watermarking for Indonesian LLM outputs.
  76. Indonesian Pronunciation Lexicon Generation Using G2P Models with Transfer Learning — Gap: Low-resource speech (arXiv 2601.03684); grapheme-to-phoneme for Indonesian languages.
  77. Unsupervised Indonesian Morphological Segmentation Using Character-Level Transformers — Gap: Indonesian morphology (arXiv 2304.02746); unsupervised segmentation for agglutinative regional languages.
  78. Indonesian Emotion-Cause Pair Extraction in Social Media Text Using RoBERTa — Gap: Emotion analysis (arXiv 2005.11882, 2003.07723); emotion-cause in Indonesian.
  79. Indonesian Event Extraction from News Using Question Answering with Fine-Tuned LLMs — Gap: QA-based NER (arXiv 1908.09138) for event extraction; Indonesian news events.
  80. End-to-End Indonesian Speech Translation Using Whisper Fine-Tuned with LoRA — Gap: LoRA (arXiv 2501.19389) for speech tasks; Indonesian speech translation.