research area Methods and Algorithmsconference SIGGRAPH Asia
content type paperpublished August 2020

Enhanced Direct Delta Mush

AuthorsSerguei Kalentchouk, Michael Hutchinson, Deepak Tolani

View publication

Direct Delta Mush is a novel skinning deformation technique introduced by Le and Lewis (2019). It generalizes the iterative Delta Mush algorithm of Mancewicz et al (2014), providing a direct solution with improved efficiency and control. Compared to Linear Blend Skinning, Direct Delta Mush offers better quality of deformations and ease of authoring at comparable performance. However, Direct Delta Mush does not handle non-rigid joint transformations correctly which limits its application for most production environments. This paper presents an extension to Direct Delta Mush that integrates the non-rigid part of joint transformations into the algorithm. In addition, the paper also describes practical considerations for computing the orthogonal component of the transformation and stability issues observed during the implementation and testing.

Related readings and updates.

Sparse Autoencoders Are Capable LLM Jailbreak Mitigators

July 2, 2026research area Speech and Natural Language ProcessingWorkshop at ICML

Jailbreak attacks remain a persistent threat to large language model safety. We propose Context-Conditioned Delta Steering (CC-Delta), an SAE-based defense that identifies jailbreak-relevant sparse features by comparing token-level representations of the same harmful request with and without jailbreak context. Using paired harmful/jailbreak prompts, CC-Delta selects features via statistical testing and applies inference-time mean-shift steering…

Read more

Federated Learning for Speech Recognition: Revisiting Current Trends Towards Large-Scale ASR

November 30, 2023research area Privacy, research area Speech and Natural Language Processingconference NeurIPS

This paper was accepted at the Federated Learning in the Age of Foundation Models workshop at NeurIPS 2023.

While automatic speech recognition (ASR) has witnessed remarkable achievements in recent years, it has not garnered a widespread focus within the federated learning (FL) and differential privacy (DP) communities. Meanwhile, ASR is also a well suited benchmark for FL and DP as there is (i) a natural data split across users by using speaker…

Read more