Domain Multi-Speaker Text-to-Speech, Speech Synthesis

Task Generate Mel-spectrograms from text inputs

Work Separates low- and high-frequency regions and refines their representations using a dual-stage architecture to improve spectral reconstruction quality

Reference Effective Mel-Spectrogram Synthesis for Multi-Speaker Text-to-Speech with Band-Specific Acoustic Modeling, IEEE Access, 2026

Domain Multimodal Learning, Video Understanding

Task Audio-Visual Event Localization

Work Proposes a CLIP-based global context regulation mechanism that reduces event overconfidence by exploiting cross-modal similarity between audio and visual modalities

Reference Effective Audio-Visual Event Localization using CLIP-based Global Context Regulation for Mitigating Event Overconfidence, Pattern Recognition, 2026

Domain Computer Vision, Generative AI

Task Diffusion-based Image Style Transfer

Work Proposes a diffusion-based style transfer framework that blends visual and textual embeddings from both source and reference images to maintain structural consistency while effectively applying artistic styles

Reference Blended Embedding Guided Style Transfer in Inversion-Based Diffusion for Creatively-Matched Source-Reference Pairs, Neurocomputing, 2026

Domain Generative AI, Background Animation Generation, Text-to-Image & Text-to-Video Models, Children's Content Creation

Task Automatically create background animations aligned with children's song lyrics

Work Modular pipeline using LLMs, diffusion models, and dynamic effects to generate children's song animations

Reference Automatic Background Animation Generation Aligned with LLM-Generated Lyrics for Children's Songs, Scientific Reports, 2026

Domain Computer Vision, Vehicle Damage Segmentatione

Task Semantic segmentation of damage regions with multiple instances, scale variation, and irregular boundaries

Work Proposes an encoder–decoder network tailored for multiple multi-scale jagged masks, enabling accurate segmentation of complex vehicle damage patterns

Reference Effective Encoder-Decoder Network for Multiple Multi-Scale Jagged Masks in Vehicle Damage Segmentation, IEEE Access, 2026

Domain Data Augmentation, Legal Domain Models

Task Legal Case Classification

Work Proposes a TF-DF–based token masking augmentation that preserves legally critical terminology and empirically validates its effectiveness across transformer models in large-scale legal case classification

Reference Effective Token Masking Augmentation Using Term-Document Frequency For Language Model-Based Legal Case Classification, CMC-Computers, Materials & Continua, 2025

Domain Crack Segmentation, Lightweight Neural Network, Receptive Field Expansion

Task Crack segmentation with balanced accuracy–efficiency trade-off

Work Lightweight crack segmentation architecture that integrates multi-scale dilated blocks with an input-guided gate to preserve fine evidence while expanding the receptive field

Reference EP-REx: Evidence-Preserving Receptive-Field Expansion for Efficient Crack Segmentation, Symmetry, 2025

Domain Computer Vision, Industrial Anomaly Detection, Unsupervised Learning

Task Literature Survey

Work Reviews state-of-the-art anomaly detection methods and conduct empirical validation for identifying the detailed drawbacks of those methods

Reference Survey on 3D Anomaly Detection Regarding Defect Size: Tradeoff Between Accuracy and Efficiency with Future Research Direction, IEEE Access, 2025

Domain Computer Vision, Style Transfer

Task Transform portraits into specific artistic styles

Work Diffusion-based generative models for webtoon-style portrait stylization designed for real-world scenarios, with a focus on demo booth platforms

Reference End-to-End Design of Webtoon-Style Portrait Stylization System for Real-World Demo Booth, IEEE Access, 2025

Domain Natural Language Processing, Passage Retrieval

Task Improve retrieval accuracy by selectively deleting redundant tokens in positive passages

Work Proposes a frequency-based token deletion method using corpus and passage statistics for data augmentation in DPR, and empirically validates its effectiveness on QA datasets (WikiQA, NQ)

Reference Data Augmentation for Dense Passage Retrieval using Corpus-Passage Frequency-Based Token Deletion, Journal of Big Data, 2025

Domain Cyber Threat Intelligence, Phishing Text, Visual Perturbation Attack, Homoglyph Restoration

Task Restore texts visually perturbed by homoglyph characters

Work Devise a hybrid system consisting of an OCR mechanism and masked character BERT

Reference OCR-Assisted Masked BERT for Homoglyph Restoration towards Multiple Phishing Text Downstream Tasks, Computers, Materials & Continua, 2025

Domain Natural Language Processing, Cyber Security, Cyber Threat Intelligence

Task Restoration of visually perturbed texts

Work Release BitAbuse dataset, which includes real-world phishing cases

Reference BitAbuse: A Dataset of Visually Perturbed Texts for Defending Phishing Attacks, NAACL, 2025

Domain Information Retrieval, Video Recommendation

Task Micro-Video Recommendation (YouTube Shorts)

Work Dual-graph-based micro-video recommendation neural network containing a dual-level positive graph from skip behavior interactions

Reference Exploiting Fine-Grained Skip Behaviors for Micro-video Recommendation, AAAI, 2025

Domain Tabular Learning, Market Share Prediction, Pre-release Forecasting

Task Implement an end-to-end system that predicts and reports the expected daily market share of upcoming movies before their release

Work Devise an end-to-end system to predict and visualize the pre-release daily market share forecasts for individual movies

Reference Development of Movie Market Share Prediction System and End-to-End Machine Learning Pipeline, Industrial Project, 2024

Domain Music Information Retrieval, Music Recommendation, Music Skip Prediction

Task Predict whether user skips occur in upcoming songs during a playlist streaming session

Work Music skip prediction network utilizing a twice-split 1D convolution block, reducing session noise before fusion

Reference Effective Music Skip Prediction based on Late Fusion Architecture for User-Interaction Noise, Expert Systems with Applications, 2024

Domain Music Information Retrieval, Music Genre Classification

Task Literature Survey

Work Review convolutional neural networks-based music genre classification method using spectral features

Reference A Short Survey and Comparison of CNN-based Music Genre Classification using Multiple Spectral Features, IEEE Access, 2024

Domain Computer Vision, Industrial Image Segmentation, Length and Thickness Measurement

Task Implement an end-to-end system that measures and report the length and thickness of sealing paste

Work Devise the end-to-end system that visualize the length and thickness of sealing paste on the car frame

Reference Development of Sealing Inspection System of Car Frame and End-to-End Machine Learning Pipeline, Industrial Project, 2024

Domain Natural Language Processing, Hypernym Discovery

Task Find suitable instances for a given hyponym from a predefined hypernym vocabulary

Work Introduce a pre-trained language model, Hypert, which is further pretrained using masked language modeling with Hearst pattern sentences

Reference Hypert: Hypernymy-aware BERT with Hearst Pattern Exploitation for Hypernym Discovery, Journal of Big Data, 2023

Domain Finance Information Retrieval, Stock Price Prediction

Task Predict the next day’s stock up and down direction using the previous ten days of data

Work Early fusion neural network consisting of feature concatenation, a transformer encoder, and a classifier layer

Reference Effective Exploitation of Macroeconomic Indicators for Stock Direction Classification using the Multimodal Fusion Transformer, IEEE Access, 2023

Domain Computer Vision, Object Detection, Multitask Learning, Embedded AI

Task Implement simultaneous pedestrian detection and text recognition on embedded board

Work Devise a multi-task neural network considering the hardware specification of the Rockchip board

Reference Development of simultaneous pedestrian detection and text recognition on embedded board, Industrial Project, 2022

Domain Music Information Retrieval, Music Genre Classification, Feature Selection

Task Literature Survey

Work Reviews feature selection method effective to the genre classification accuracy

Reference Toward a Fair Evaluation and Analysis of Feature Selection for Music Tag Classification, IEEE Access, 2021

Domain Educational Data Mining, Computer-Aided Pronunciation Training, Word Recommendation

Task Recommend a set of words that helps improve users’ pronunciation skills

Work Propose a mobile CAPT system targeting unacceptable pronunciation due to confusion among contextual allophones

Reference Effective Computer-Assisted Pronunciation Training Based on Phone-Sensitive Word Recommendation, Concurrency and Computation: Practice and Experience, 2021

Domain Robot Path Planning, Coverage Path Planning, Robotic Vacuum Cleaner

Task Identify the longest path in the given territory for the robotic vacuum cleaner

Work Propose a novel map decomposition method for creating the longest path efficiently

Reference Scalable Coverage Path Planning for Cleaning Robots using Rectangular Map Decomposition on Large Environments, IEEE Access, 2018

Domain Educational Data Mining, Computer-Aided Pronunciation Training, Word Recommendation

Task Recommend a set of words that helps improve users’ pronunciation skills

Work Propose a mobile CAPT system recommending a set of words based on the weights of the phonemes in the tested words

Reference Smartphone-Assisted Pronunciation Learning Technique for Ambient Intelligence, IEEE Access, 2017

Domain Music Information Retrieval, Music Search, Tag Combination

Task Identify a music clip that the user targeted as fast as possible

Work Propose a music retrieval system based on multiple (in)dependent tags

Reference Effective Music Searching Approach based on Tag-Combination by exploiting Prototypical Acoustic Content, Multimedia Tools and Applications, 2017

Domain Robot Path Planning, Shortest Path Planning

Task Identify the shortest path from the start to the destination point without interfering obstacles

Work Devise an effective initialization method based on the rapidly-exploring random tree for metaheuristic path planning methods

Reference An effective initialization method for genetic algorithm-based robot path planning using a directed acyclic graph, Information Sciences, 2016