Multi-Loss Weighting with Coefficient of Variations

Groenendijk, Rick; Karaoğlu, Sezer; Gevers, Theo; Mensink, Thomas

doi:10.48550/arxiv.2009.01717

Cited by 2 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hence, it accounts for the dynamic range of loss terms in a principled way. Coefficient of Variation [9] states that a loss term is satisfied when its variance has decreased to zero. Hence, it assigns weight for a loss term as its coefficient of variation, i.e., standard deviation divided by mean.…”

Section: Brief Overview Of Multi-task Learning Methodsmentioning

confidence: 99%

“…Note that this loss is different from usual formulations of transfer learning, knowledge distillation, and even multi-task learning. [9] appropriately terms such setting as singletask multi-loss. Our framework is flexible and, instead of speech denoising, it can be used for voice conversion and domain adaptation as well.…”

Section: Speech Denoising With Perceptual Ensemble Regularization Los...mentioning

confidence: 99%

See 1 more Smart Citation

Perceptual Loss Based Speech Denoising with an Ensemble of Audio Pattern Recognition and Self-Supervised Models

Kataria

Villalba

Dehak

2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

Deep learning based speech denoising still suffers from the challenge of improving perceptual quality of enhanced signals. We introduce a generalized framework called Perceptual Ensemble Regularization Loss (PERL) built on the idea of perceptual losses. Perceptual loss discourages distortion to certain speech properties and we analyze it using six large-scale pre-trained models: speaker classification, acoustic model, speaker embedding, emotion classification, and two self-supervised speech encoders (PASE+, wav2vec 2.0). We first build a strong baseline (w/o PERL) using Conformer Transformer Networks on the popular enhancement benchmark called VCTK-DEMAND. Using auxiliary models one at a time, we find acoustic event and self-supervised model PASE+ to be most effective. Our best model (PERL-AE) only uses acoustic event model (utilizing AudioSet) to outperform state-of-the-art methods on major perceptual metrics. To explore if denoising can leverage full framework, we use all networks but find that our seven-loss formulation suffers from the challenges of Multi-Task Learning. Finally, we report a critical observation that state-of-the-art Multi-Task weight learning methods cannot outperform hand tuning, perhaps due to challenges of domain mismatch and weak complementarity of losses.

show abstract

Section: Brief Overview Of Multi-task Learning Methodsmentioning

confidence: 99%

Section: Speech Denoising With Perceptual Ensemble Regularization Los...mentioning

confidence: 99%

Perceptual Loss Based Speech Denoising with an Ensemble of Audio Pattern Recognition and Self-Supervised Models

Kataria

Villalba

Dehak

2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

show abstract

“…The Coefficient-of-Variations (CoV) weighting scheme is also designed for optimally learning machine learning tasks where the objective function is a weighted linear combination of multiple losses. 56 In brief, this weighting scheme uses the statistical properties inherent in loss functions to explicitly determine their relative importance. This is achieved using the CoV, which quantifies the relationship between the standard deviation (σ) and the mean (µ).…”

Section: Coefficient-of-variations Weightingmentioning

confidence: 99%

Structural Coarse-Graining via Multi-Objective Optimization with Differentiable Simulation

Wu,

Zhou

2023

Preprint

View full text Add to dashboard Cite

In the realm of multiscale molecular simulations, structure-based coarse graining is a prominent approach for creating efficient coarse-grained (CG) representations of soft matter systems such as polymers. This involves optimizing CG interactions by matching static correlation functions of corresponding degrees of freedom in all-atom (AA) models. Here, we present a versatile method, namely, differentiable coarse-graining (DiffCG), which combines multi-objective optimization and differentiable simulation. The DiffCG approach is capable of constructing robust CG models by iteratively optimizing effective potentials to simultaneously match multiple target properties. We demonstrate our approach by concurrently optimizing bonded and non-bonded potentials of a CG model of polystyrene (PS) melts. The resulting CG-PS model accurately reproduces both structural and thermodynamic properties of the AA counterpart. More importantly, leveraging the multi-objective optimization capability, we develop a precise and efficient CG model for PS melts that is transferable across a wide range of temperatures, i.e., from $400$ to $600$ K. It is achieved via optimizing a pairwise potential with nonlinear temperature dependence in the CG model to simultaneously match target data from AA-MD simulations at multiple thermodynamic states. Our work showcases a promising route for developing accurate and transferable CG models of complex soft-matter systems through multi-objective optimization with differentiable simulation.

show abstract

Multi-Loss Weighting with Coefficient of Variations

Cited by 2 publications

References 0 publications

Perceptual Loss Based Speech Denoising with an Ensemble of Audio Pattern Recognition and Self-Supervised Models

Perceptual Loss Based Speech Denoising with an Ensemble of Audio Pattern Recognition and Self-Supervised Models

Structural Coarse-Graining via Multi-Objective Optimization with Differentiable Simulation

Contact Info

Product

Resources

About