Researchers are developing new methods for unsupervised cross-modal retrieval, aiming to improve efficiency and reduce reliance on large, manually annotated datasets. Papers propose techniques like Attribute-Prompted Kernel Hashing (APKH) and Global-Neighborhood Alignment Hashing (GNAH) that leverage vision-language foundation models and limited paired data to construct compact, aligned Hamming spaces. Another approach, UniCA, introduces bi-directional cross-attention and a positive similarity loss for more robust multi-modal retrieval, demonstrating improvements on benchmarks like WebQA+. AI
IMPACT These research efforts aim to make cross-modal retrieval more accessible and efficient by reducing data requirements and improving alignment techniques.
RANK_REASON Multiple academic papers proposing new methods for cross-modal retrieval.
Read on arXiv cs.IR (Information Retrieval) →
- arXiv
- Bi-directional Cross-Attention
- Contrastive Loss
- cosine similarity
- information retrieval
- language model
- MSE loss
- multilayer perceptron
- Multimodal retrieval of autobiographical memories: sensory information contributes differently to the recollection of events
- Positive Similarity Loss
- transformer-based Models
- vision-language model
- WebQA+
- Attribute-Prompted Kernel Hashing
- cross-modal retrieval
- Fan Xu
- Global-Neighborhood Alignment Hashing
- unsupervised cross-modal hashing
- vision-language foundation models
AI-generated summary · Google Gemini · from 6 sources. How we write summaries →
