Page Not Found
Page not found. Your pixels are in another canvas.
Sitemap
A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.
Pages
Welcome!
Dylan Paiton
Page not in menu
This is a page not in th emain menu
Posts
Response geometry explains adversarial robustness
Published:
Defending against adversarial attacks is critical for modern AI applications. Here I describe how we can use response geometry to predict a neuron’s susceptibility to attacks.
Understanding neurons using response geometry
Published:
Scientists have long summarized neurons in terms of the relationship between their inputs and outputs. Here I describe a technique that allows us to go beyond previous approaches to succinctly describe important and complex neural behavior.
publications
Combining multiple visual processing streams for locating and classifying objects in video
Published in IEEE Southwest Symposium on Image Analysis and Interpretation, 2012
Labeling videos with objects using independent color/texture and shape/form processing streams.
Biologically inspired distributed sensor networks: Collective signal amplification via ultra-low bandwidth spike-based communication
Published in The International Joint Conference on Neural Networks (IJCNN), 2013
Computer simulations of distributed sensor networks using a retina-inspired communication protocol to amplify signals.
Replicating kernels with a short stride allows sparse reconstructions with fewer independent kernels
Published in arXiv Preprint, 2014
Exploring the tradeoff of patch size, stride, and overcompleteness in convolutional sparse coding.
Sparse encoding of binocular images for depth inference
Published in IEEE Southwest Symposium on Image Analysis and Interpretation (SSIAI), 2016
A convolutional sparse coding network facilitates better depth inference than comparable feedforward networks.
A deconvolutional competitive algorithm for building sparse hierarchical representations
Published in Proceedings of the 9th EAI International Conference on Bio-inspired Information and Communications Technologies, 2016
A hierarchical sparse coding network that learns bandpass decompositions of natural images.
Joint source-channel coding with neural networks for analog data compression and storage
Published in Data Compression Conference, 2018
A convolutional autoencoder with divisive normalization enables digital image storage on simulated emerging memristive devices.
The sparse manifold transform
Published in Neural Information Processing Systems, 2018
A hierarchical sparse coding model that decomposes scenes into constituent parts and linearizes temporal trajectories of natural videos.
Error-resilient analog image storage and compression with analog-valued RRAM arrays: an adaptive joint source-channel coding approach
Published in IEEE International Electron Devices Meeting (IEDM), 2018
The first published approach for storing digital natural images onto resistive random access memory arrays.
Analysis and applications of the locally competitive algorithm
Published in UC Berkeley Thesis, 2019
My PhD thesis provides an in-depth account of a recurrent network for sparse inference, including novel analyses, comparisons, and extensions.
Subspace locally competitive algorithms
Published in Neuro-Inspired Computational Elements, 2020
We present a 2-layer recurrent sparse coding network for learning higher order statistical regularities in natural images.
Selectivity and robustness of sparse coding networks
Published in Journal of Vision, 2020
Using differential geometry we explain how sparse coding networks bend their response surfaces, which results in improved selectivity and robustness for individual neurons.
Towards nonlinear disentanglement in natural data with temporal sparse coding
Published in International Conference on Learning Representations, 2021
We construct and analyze new datasets for evaluating disentanglement on natural videos. We also propose a temporally sparse prior for identifying the underlying factors of variation in natural videos.
research
Neuron Response Geometry
Published:
Understanding neural computation using differential geometry
Hierarchical models of natural scenes
Published:
Proposing candidate models for biological scene representation
talks
Towards Nonlinear Disentanglement in Natural Data with Temporal Sparse Coding
Published:
We construct an unsupervised learning model that achieves nonlinear disentanglement of underlying factors of variation in naturalistic videos. Previous work suggests that representations can be disentangled if all but a few factors in the environment stay constant at any point in time. As a result, algorithms proposed for this problem have only been tested on carefully constructed datasets with this exact property, leaving it unclear whether they will transfer to natural scenes. Here we provide evidence that objects in segmented natural movies undergo transitions that are typically small in magnitude with occasional large jumps, which is characteristic of a temporally sparse distribution. We leverage this finding and present SlowVAE, a model for unsupervised representation learning that uses a sparse prior on temporally adjacent observations to disentangle generative factors without any assumptions on the number of changing factors. We provide a proof of identifiability and show that the model reliably learns disentangled representations on several established benchmark datasets, often surpassing the current state-of-the-art. We additionally demonstrate transferability towards video datasets with natural dynamics, Natural Sprites and KITTI Masks, which we contribute as benchmarks for guiding disentanglement research towards more natural data domains.
Subspace Locally Competitive Algorithms
Published:
We introduce subspace locally competitive algorithms (SLCAs), a family of novel network architectures for modeling latent representations of natural signals with group sparse structure. SLCA first layer neurons are derived from locally competitive algorithms, which produce responses and learn representations that are well matched to both the linear and non-linear properties observed in simple cells in layer 4 of primary visual cortex (area V1). SLCA incorporates a second layer of neurons which produce approximately invariant responses to signal variations that are linear in their corresponding subspaces, such as phase shifts, resembling a hallmark characteristic of complex cells in V1. We provide a practical analysis of training parameter settings, explore the features and invariances learned, and finally compare the model to single-layer sparse coding and to independent subspace analysis.
teaching
Teaching experience 1
Undergraduate course, University 1, Department, 2014
This is a description of a teaching experience. You can use markdown like any other post.
Teaching experience 2
Workshop, University 1, Department, 2015
This is a description of a teaching experience. You can use markdown like any other post.