Bahareh Tolooshams
@btolooshams.bsky.social
Assistant Professor @ualberta.bsky.social | Postdoc @caltech.edu | PhD from @harvard.edu | https://btolooshams.github.io
Our preliminary results show that EquiReg can also be used to improve diffusion-based text-to-image guidance and generate images that are more natural-looking and feasible (see the paper for more visualizations). 14/
June 12, 2025 at 3:47 PM
Our preliminary results show that EquiReg can also be used to improve diffusion-based text-to-image guidance and generate images that are more natural-looking and feasible (see the paper for more visualizations). 14/
Improved PDE reconstruction performance: EquiReg enhances performance on PDE solving tasks. For instance, Equi-FunDPS reduces the error of FunDPS on Helmholtz and Navier-Stokes inverse problems. 13/
June 12, 2025 at 3:47 PM
Improved PDE reconstruction performance: EquiReg enhances performance on PDE solving tasks. For instance, Equi-FunDPS reduces the error of FunDPS on Helmholtz and Navier-Stokes inverse problems. 13/
EquiReg improves the perceptual quality in image restoration tasks. 12/
June 12, 2025 at 3:47 PM
EquiReg improves the perceptual quality in image restoration tasks. 12/
Improved image restoration performance: We demonstrate that EquiReg significantly improves performance on linear and nonlinear image restoration tasks. EquiReg also consistently shows improvement in performance across many noise levels. 11/
June 12, 2025 at 3:47 PM
Improved image restoration performance: We demonstrate that EquiReg significantly improves performance on linear and nonlinear image restoration tasks. EquiReg also consistently shows improvement in performance across many noise levels. 11/
We present two strategies for finding MPE functions:
a) training induced: equivariance emerges in encoder-decoder architectures trained with symmetry-preserving augmentations,
b) data inherent: MPE arises from inherent data symmetries, commonly observed in physical systems. 9/
a) training induced: equivariance emerges in encoder-decoder architectures trained with symmetry-preserving augmentations,
b) data inherent: MPE arises from inherent data symmetries, commonly observed in physical systems. 9/
June 12, 2025 at 3:47 PM
We present two strategies for finding MPE functions:
a) training induced: equivariance emerges in encoder-decoder architectures trained with symmetry-preserving augmentations,
b) data inherent: MPE arises from inherent data symmetries, commonly observed in physical systems. 9/
a) training induced: equivariance emerges in encoder-decoder architectures trained with symmetry-preserving augmentations,
b) data inherent: MPE arises from inherent data symmetries, commonly observed in physical systems. 9/
We introduce Manifold-Preferential Equivariant (MPE) functions, which exhibit low equivariance error on the support of the data manifold (in-distribution) and higher error off-manifold (out-of-distribution). Our proposed regularization, EquiReg, is based on this MPE property. 8/
June 12, 2025 at 3:47 PM
We introduce Manifold-Preferential Equivariant (MPE) functions, which exhibit low equivariance error on the support of the data manifold (in-distribution) and higher error off-manifold (out-of-distribution). Our proposed regularization, EquiReg, is based on this MPE property. 8/
EquiReg with distribution-dependent equivariance errors: We propose a new class of regularizers grounded in distribution-dependent equivariance error, a formalism that quantifies how symmetry violations vary depending on whether samples lie on- or off-manifold. 7/
June 12, 2025 at 3:47 PM
EquiReg with distribution-dependent equivariance errors: We propose a new class of regularizers grounded in distribution-dependent equivariance error, a formalism that quantifies how symmetry violations vary depending on whether samples lie on- or off-manifold. 7/
This addresses errors arising in the likelihood score due to poor posterior approx and prior score errors resulting from off-manifold trajectories. To realize such a regularizer, we seek an approach for manifold alignment via global properties of the data distribution. 5/
June 12, 2025 at 3:47 PM
This addresses errors arising in the likelihood score due to poor posterior approx and prior score errors resulting from off-manifold trajectories. To realize such a regularizer, we seek an approach for manifold alignment via global properties of the data distribution. 5/
This breaks down for highly complex distributions: Conditional expectation, as in the posterior mean expectation, computes a linear combination of all possible x0; hence, from the manifold perspective, the posterior mean expectation may end up being located off-manifold. 3/
June 12, 2025 at 3:47 PM
This breaks down for highly complex distributions: Conditional expectation, as in the posterior mean expectation, computes a linear combination of all possible x0; hence, from the manifold perspective, the posterior mean expectation may end up being located off-manifold. 3/
🚨We propose EquiReg, a generalized regularization framework that uses symmetry in generative diffusion models to improve solutions to inverse problems. arxiv.org/abs/2505.22973
@aditijc.bsky.social, Rayhan Zirvi, Abbas Mammadov, @jiacheny.bsky.social, Chuwei Wang, @anima-anandkumar.bsky.social 1/
@aditijc.bsky.social, Rayhan Zirvi, Abbas Mammadov, @jiacheny.bsky.social, Chuwei Wang, @anima-anandkumar.bsky.social 1/
June 12, 2025 at 3:47 PM
🚨We propose EquiReg, a generalized regularization framework that uses symmetry in generative diffusion models to improve solutions to inverse problems. arxiv.org/abs/2505.22973
@aditijc.bsky.social, Rayhan Zirvi, Abbas Mammadov, @jiacheny.bsky.social, Chuwei Wang, @anima-anandkumar.bsky.social 1/
@aditijc.bsky.social, Rayhan Zirvi, Abbas Mammadov, @jiacheny.bsky.social, Chuwei Wang, @anima-anandkumar.bsky.social 1/
Our decoding package also offers a movement-imitated augmentation framework (VARS-fUSI++). By augmenting the image for decoder training with small, randomly rotated and translated images, you can increase the decoder's robustness, hence, performance. 12/
April 28, 2025 at 5:55 PM
Our decoding package also offers a movement-imitated augmentation framework (VARS-fUSI++). By augmenting the image for decoder training with small, randomly rotated and translated images, you can increase the decoder's robustness, hence, performance. 12/
We show that VARS-fUSI can be generalized to human participants while maintaining decodeable behavior-correlated information, generating decoding accuracies and activation maps comparable to ground truth. 11/
April 28, 2025 at 5:55 PM
We show that VARS-fUSI can be generalized to human participants while maintaining decodeable behavior-correlated information, generating decoding accuracies and activation maps comparable to ground truth. 11/
In a motor decoding experiment, the direction of planned saccadic eye movements can be decoded as accurately from VARS-fUSI as from ground truth. The decoding relied on the same region of posterior parietal cortex, showing that VARS-fUSI conserves behavioral information. 10/
April 28, 2025 at 5:55 PM
In a motor decoding experiment, the direction of planned saccadic eye movements can be decoded as accurately from VARS-fUSI as from ground truth. The decoding relied on the same region of posterior parietal cortex, showing that VARS-fUSI conserves behavioral information. 10/
Despite the extensive reduction of required pulses, VARS-fUSI can still be used reliably for low-latency and efficient behavioral decoding in brain-computer interfaces (BCIs). We demonstrate this in monkey and human experiments. 9/
April 28, 2025 at 5:55 PM
Despite the extensive reduction of required pulses, VARS-fUSI can still be used reliably for low-latency and efficient behavioral decoding in brain-computer interfaces (BCIs). We demonstrate this in monkey and human experiments. 9/
Why does VARS-fUSI perform better than other AI models?
- It respects the complex-valued nature of the data,
- It treats the temporal aspect of the data as a continuous function using neural operators,
- It captures decomposed spatially local and temporally global features. 8/
- It respects the complex-valued nature of the data,
- It treats the temporal aspect of the data as a continuous function using neural operators,
- It captures decomposed spatially local and temporally global features. 8/
April 28, 2025 at 5:55 PM
Why does VARS-fUSI perform better than other AI models?
- It respects the complex-valued nature of the data,
- It treats the temporal aspect of the data as a continuous function using neural operators,
- It captures decomposed spatially local and temporally global features. 8/
- It respects the complex-valued nature of the data,
- It treats the temporal aspect of the data as a continuous function using neural operators,
- It captures decomposed spatially local and temporally global features. 8/
What does VARS-fUSI offer to practitioners using fUSI in clinical or neuroscience research?
VARS-fUSI offers a training pipeline, a fine-tuning procedure, and an accelerated, low-latency online imaging system. You can use VARS-fUSI to reduce the acquisition duration or rate. 7/
VARS-fUSI offers a training pipeline, a fine-tuning procedure, and an accelerated, low-latency online imaging system. You can use VARS-fUSI to reduce the acquisition duration or rate. 7/
April 28, 2025 at 5:55 PM
What does VARS-fUSI offer to practitioners using fUSI in clinical or neuroscience research?
VARS-fUSI offers a training pipeline, a fine-tuning procedure, and an accelerated, low-latency online imaging system. You can use VARS-fUSI to reduce the acquisition duration or rate. 7/
VARS-fUSI offers a training pipeline, a fine-tuning procedure, and an accelerated, low-latency online imaging system. You can use VARS-fUSI to reduce the acquisition duration or rate. 7/
VARS-fUSI is SOTA!
We applied VARS-fUSI to brain images collected from mouse, monkey, and human. Our method achieves state-of-the-art performance and shows superior generalization to unseen sessions in new animals, and even across species. 6/
We applied VARS-fUSI to brain images collected from mouse, monkey, and human. Our method achieves state-of-the-art performance and shows superior generalization to unseen sessions in new animals, and even across species. 6/
April 28, 2025 at 5:55 PM
VARS-fUSI is SOTA!
We applied VARS-fUSI to brain images collected from mouse, monkey, and human. Our method achieves state-of-the-art performance and shows superior generalization to unseen sessions in new animals, and even across species. 6/
We applied VARS-fUSI to brain images collected from mouse, monkey, and human. Our method achieves state-of-the-art performance and shows superior generalization to unseen sessions in new animals, and even across species. 6/
Change the sampling rate or duration of your data acquisition at inference time, and VARS-fUSI will still provide high-quality images. This is not possible with current AI models. VARS-fUSI reconstructs images using 10% of the time or sampling rate typically needed per image. 4/
April 28, 2025 at 5:55 PM
Change the sampling rate or duration of your data acquisition at inference time, and VARS-fUSI will still provide high-quality images. This is not possible with current AI models. VARS-fUSI reconstructs images using 10% of the time or sampling rate typically needed per image. 4/
We have released VARS-fUSI: Variable sampling for fast and efficient functional ultrasound imaging (fUSI) using neural operators.
The first deep learning fUSI method to allow for different sampling durations and rates during training and inference. biorxiv.org/content/10.1... 1/
The first deep learning fUSI method to allow for different sampling durations and rates during training and inference. biorxiv.org/content/10.1... 1/
April 28, 2025 at 5:55 PM
We have released VARS-fUSI: Variable sampling for fast and efficient functional ultrasound imaging (fUSI) using neural operators.
The first deep learning fUSI method to allow for different sampling durations and rates during training and inference. biorxiv.org/content/10.1... 1/
The first deep learning fUSI method to allow for different sampling durations and rates during training and inference. biorxiv.org/content/10.1... 1/
We plug DiffStateGrad into several SOTA diffusion solvers such as DPS, PSLD, ReSample, and DAPS. DiffStateGrad consistently improves the performance of both pixel-based and latent models. We demonstrate this on linear and nonlinear image restoration inverse problem tasks. 6/
April 24, 2025 at 4:58 AM
We plug DiffStateGrad into several SOTA diffusion solvers such as DPS, PSLD, ReSample, and DAPS. DiffStateGrad consistently improves the performance of both pixel-based and latent models. We demonstrate this on linear and nonlinear image restoration inverse problem tasks. 6/
DiffStateGrad significantly improves worst-case performance, reducing the failure rate from 26% to 4% on phase retrieval tasks when applied to ReSample. 5/
April 24, 2025 at 4:58 AM
DiffStateGrad significantly improves worst-case performance, reducing the failure rate from 26% to 4% on phase retrieval tasks when applied to ReSample. 5/
DiffStateGrad improves the robustness of diffusion-based inverse solvers to the measurement guidance step size and noise. 4/
April 24, 2025 at 4:58 AM
DiffStateGrad improves the robustness of diffusion-based inverse solvers to the measurement guidance step size and noise. 4/
DiffStateGrad uses SVD to define the subspace for gradient projection. We show that it's not the low-rank aspect of the projection but the choice of the subspace estimating the tangent space of the diffusion manifold that improves performance. 3/
April 24, 2025 at 4:58 AM
DiffStateGrad uses SVD to define the subspace for gradient projection. We show that it's not the low-rank aspect of the projection but the choice of the subspace estimating the tangent space of the diffusion manifold that improves performance. 3/
We introduce DiffStateGrad, which improves diffusion models for inverse problems. DiffStateGrad projects the measurement gradient onto a low-rank subspace of the diffusion state, improving updates stay close to the manifold and reducing artifact generation. 2/
April 24, 2025 at 4:58 AM
We introduce DiffStateGrad, which improves diffusion models for inverse problems. DiffStateGrad projects the measurement gradient onto a low-rank subspace of the diffusion state, improving updates stay close to the manifold and reducing artifact generation. 2/
How does the brain integrate prior knowledge with sensory data to perceive the world?
Come check out our poster [1-090] at #cosyne2025:
"A feedback mechanism in generative networks to remove visual degradation," joint work with Yuelin Shi, @anima-anandkumar.bsky.social, and Doris Tsao. 1/2
Come check out our poster [1-090] at #cosyne2025:
"A feedback mechanism in generative networks to remove visual degradation," joint work with Yuelin Shi, @anima-anandkumar.bsky.social, and Doris Tsao. 1/2
March 27, 2025 at 8:59 PM
How does the brain integrate prior knowledge with sensory data to perceive the world?
Come check out our poster [1-090] at #cosyne2025:
"A feedback mechanism in generative networks to remove visual degradation," joint work with Yuelin Shi, @anima-anandkumar.bsky.social, and Doris Tsao. 1/2
Come check out our poster [1-090] at #cosyne2025:
"A feedback mechanism in generative networks to remove visual degradation," joint work with Yuelin Shi, @anima-anandkumar.bsky.social, and Doris Tsao. 1/2