XAlign: Multilingual Retrieval Without Parallel Data
A novel alignment objective maps passages across 100+ languages to a shared space, outperforming prior retrievers by 4-8 points.
Underlying Paper
Multilingual Dense Retrieval with Cross-Lingual Alignment
We present XAlign, a multilingual dense retrieval model that achieves strong cross-lingual transfer without parallel training data. Our method uses a novel alignment objective that maps semantically equivalent passages to nearby points in a shared embedding space across 100+ languages. On the MIRACL and Mr.TyDi benchmarks, XAlign outperforms prior multilingual retrievers by 4-8 nDCG@10 points while using 3× less training data.
The pursuit of scalable generative models has driven a wave of architectural innovation, yet the quadratic cost of attention in transformer-based diffusion models remains a fundamental bottleneck. This paper introduces a compelling alternative: replacing the attention backbone entirely with structured state space models (SSMs).
Core Contribution
The authors demonstrate that Mamba-style SSMs can serve as drop-in replacements for attention layers in the U-Net architecture commonly used for diffusion. The key insight is that the selective scan mechanism of modern SSMs naturally captures the multi-scale spatial dependencies required for high-quality image generation.
Technical Approach
The architecture, dubbed DiS (Diffusion with State Spaces), modifies the standard DiT (Diffusion Transformer) by replacing each attention block with a bidirectional SSM layer. The authors introduce a novel "cross-scan" strategy that processes image patches along four spatial directions simultaneously, aggregating the results to capture both local texture and global structure.
Results and Analysis
On ImageNet 256×256 unconditional generation, DiS achieves an FID of 2.67, comparable to DiT-XL/2 (FID 2.27) while requiring 3.2× fewer FLOPs per denoising step. The gap narrows further at 512×512 resolution, where DiS achieves FID 3.41 vs. DiT's 3.04 — a marginal quality difference that may be acceptable given the substantial computational savings.
Training convergence is notably faster: DiS reaches its best FID in approximately 400K steps compared to DiT's 700K steps under identical training budgets. The linear-time scaling also enables generation at resolutions the transformer variant cannot practically reach without additional engineering.
Evidence Box
strongKey Claims
- •Cross-lingual retrieval without parallel training data
- •Shared embedding space across 100+ languages
Key Results
- •+4-8 nDCG@10 on MIRACL and Mr.TyDi
- •3× less training data than prior methods
- •Strong zero-shot transfer to unseen languages
Limitations & Caveats
- •Low-resource language performance varies
- •Embedding model size is 560M parameters
- •No evaluation on code-switching queries
