Machine Translation

TransVIP: groundbreaking speech-to-speech translation system preserving voice & isochrony, outperforming current state-of-the-art models!

Unlocking the speed and efficiency of Mixture-of-Expert models, this research unveils novel optimization techniques, achieving dramatic improvements in inference throughput and resource usage.

Researchers dramatically improved sign language translation by scaling up data, model size, and the number of languages, achieving state-of-the-art results.

QUEST, a novel Metropolis-Hastings sampling method, generates high-quality & diverse machine translations by using quality metrics as energy functions, overcoming limitations of likelihood-based and r…

PCformer boosts Transformer performance by using a predictor-corrector learning framework and exponential moving average coefficient learning for high-order prediction, achieving state-of-the-art resu…

SignCL, a novel contrastive learning strategy, significantly boosts gloss-free sign language translation by mitigating representation density, achieving substantial performance gains.

Fast approximation of Minimum Bayes Risk (MBR) decoding achieved using low-rank matrix completion algorithms, drastically reducing computational cost without sacrificing translation quality.

DIFFNORM boosts non-autoregressive speech-to-speech translation by normalizing speech data with a diffusion model and classifier-free guidance, achieving significant quality improvements.

Sparse Differentiable Tree Machine (sDTM) improves compositional generalization in neural networks by efficiently representing tree structures in vector space, enabling simultaneous symbolic and neura…

Code Rosetta pushes the boundaries of unsupervised code translation by introducing the first encoder-decoder model that efficiently translates between programming languages and their parallel HPC exte…

Revolutionizing Transformers, Algebraic Positional Encodings (APE) offers a theory-first approach to positional encoding, outperforming state-of-the-art methods without hyperparameter tuning across va…