Filters








253,878 Hits in 5.0 sec

Transformation Properties of Learned Visual Representations [article]

Taco S. Cohen, Max Welling
2015 arXiv   pre-print
Starting with the idea that a good visual representation is one that transforms linearly under scene motions, we show, using the theory of group representations, that any such representation is equivalent  ...  When a three-dimensional object moves relative to an observer, a change occurs on the observer's image plane and in the visual representation computed by a learned model.  ...  Both of these problems require us to take a closer look at the transformation properties of learned visual representations.  ... 
arXiv:1412.7659v3 fatcat:orjfnrh6hvbvtbicaormxs453e

Self-supervised Wide Baseline Visual Servoing via 3D Equivariance [article]

Jinwook Huh, Jungseok Hong, Suveer Garg, Hyun Soo Park, Volkan Isler
2022 arXiv   pre-print
We learn a coherent visual representation by leveraging a geometric property called 3D equivariance-the representation is transformed in a predictable way as a function of 3D transformation.  ...  With the learned model, the relative transformation can be inferred simply by following the gradient in the learned space and used as feedback for closed-loop visual servoing.  ...  Specifically, there exists a visual representation that possesses a geometric property called 3D equivariancethe visual representation is transformed as a function of 3D transformation.  ... 
arXiv:2209.05432v1 fatcat:invlqkinajbt3ntxvhpqz3tm5a

Learning Invariant Object and Spatial View Representations in the Brain Using Slow Unsupervised Learning

Edmund T. Rolls
2021 Frontiers in Computational Neuroscience  
For the ventral visual system, one key adaptation is the use of information available in the statistics of the environment in slow unsupervised learning to learn transform-invariant representations of  ...  First, neurophysiological evidence for the learning of invariant representations in the inferior temporal visual cortex is described.  ...  FIGURE 8 | 8 FIGURE 8 | Continuous spatial transformation learning of transform-invariant visual representations of objects.  ... 
doi:10.3389/fncom.2021.686239 fatcat:xbthyagwyjb57c7f6ajkxl2diy

Learning invariant and variant components of time-varying natural images

B. Olshausen, C. Cadieu
2010 Journal of Vision  
A remarkable property of biological visual systems is their ability to infer structure within the visual world.  ...  How do biological systems decompose visual information into separate invariant and variant representations?  ...  A remarkable property of biological visual systems is their ability to infer structure within the visual world.  ... 
doi:10.1167/7.9.964 fatcat:sfjc3bpvbjgr5i3bxussclipye

AB009. Learning dynamics in a neural network model of the primary visual cortex

Hugo Ladret, Laurent Perrinet
2019 Annals of Eye Science  
biorealism in the context of learning visual inputs.  ...  The primary visual cortex (V1) is a key component of the visual system that builds some of the first levels of coherent visual representations from sparse visual inputs.  ...  Background: The primary visual cortex (V1) is a key component of the visual system that builds some of the first levels of coherent visual representations from sparse visual inputs.  ... 
doi:10.21037/aes.2019.ab009 fatcat:j7zxq2odzna3xmieyufruqggqq

Invariant Recognition Shapes Neural Representations of Visual Input

Andrea Tacchetti, Leyla Isik, Tomaso A. Poggio
2018 Annual Review of Vision Science  
in the visual world shapes which neural representations of sensory input are computed by human visual cortex. 403 Annu.  ...  The ability to generalize across these complex transformations is a hallmark of human visual intelligence, which has been the focus of wide-ranging investigation in systems and computational neuroscience  ...  main AL properties, including mirror symmetry, have to be understood as transformations of ML representations and the main AM properties as transformations of AL representations (Freiwald & Tsao 2010)  ... 
doi:10.1146/annurev-vision-091517-034103 pmid:30052494 fatcat:hztby7443jgzzp5mjvmfc7coou

Symmetry-Based Representations for Artificial and Biological General Intelligence

Irina Higgins, Sébastien Racanière, Danilo Rezende
2022 Frontiers in Computational Neuroscience  
Finally, first demonstrations of the importance of symmetry transformations for representation learning in the brain are starting to arise in neuroscience.  ...  It is believed that learning "good" sensory representations is important for enabling this, however there is little agreement as to what a good representation should look like.  ...  (A)The properties of the visual data obtained through a head camera from toddlers (Smith et al., 2018; Slone et al., 2019) is similar to the properties of the visual data that allows ML approaches to  ... 
doi:10.3389/fncom.2022.836498 pmid:35493854 pmcid:PMC9049963 fatcat:qpj4okhbojat5p3v4smcqspizq

ChemBERTa: Large-Scale Self-Supervised Pretraining for Molecular Property Prediction [article]

Seyone Chithrananda, Gabriel Grand, Bharath Ramsundar
2020 arXiv   pre-print
Our results suggest that transformers offer a promising avenue of future work for molecular representation learning and property prediction.  ...  However, in NLP, transformers have become the de-facto standard for representation learning thanks to their strong downstream task transfer.  ...  on molecular strings, as well as in motivating the utilization of SELFIES in this work.  ... 
arXiv:2010.09885v2 fatcat:dpt4gwxccngg5nbyp6pumfolra

Symmetry-Based Representations for Artificial and Biological General Intelligence [article]

Irina Higgins, Sébastien Racanière, Danilo Rezende
2022 arXiv   pre-print
Finally, first demonstrations of the importance of symmetry transformations for representation learning in the brain are starting to arise in neuroscience.  ...  It is believed that learning "good" sensory representations is important for enabling this, however there is little agreement as to what a good representation should look like.  ...  A: The properties of the visual data obtained through a head camera from toddlers (Slone et al., 2019; Smith et al., 2018) is similar to the properties of the visual data that allows ML approaches to  ... 
arXiv:2203.09250v1 fatcat:fgkcxnndnraqjpg4sb4pflvha4

Probing Representations Learned by Multimodal Recurrent and Transformer Models [article]

Jindřich Libovický, Pranava Madhyastha
2019 arXiv   pre-print
It also has been shown that visual information serves as one of the means for grounding sentence representations.  ...  We evaluate textual and visual features of sentence representations obtained using predominant approaches on image retrieval and semantic textual similarity.  ...  Experiments with multimodal language models (LMs) also confirm that multimodality influences the semantic properties of learned representations (Poerner et al., 2018) .  ... 
arXiv:1908.11125v1 fatcat:b5ktvnumqverbouczrspqgcwo4

Invariant visual object recognition: biologically plausible approaches

Leigh Robinson, Edmund T. Rolls
2015 Biological cybernetics  
We also acknowledge the use of Blender software (http://www.blender.org) to render the 3D objects, of the CalTech-256 (Griffin et al. 2007 ), and of the Amsterdam Library of Images (ALOI) (Geusebroek  ...  Acknowledgments The use of the ORL database of faces http://www. cl.cam.ac.uk/research/dtg/attarchive/facedatabase.html) provided by AT&T Laboratories Cambridge is acknowledged.  ...  VisNet can learn to respond to the different transforms of objects using the trace learning rule to capture the properties of objects as they transform in the world.  ... 
doi:10.1007/s00422-015-0658-2 pmid:26335743 pmcid:PMC4572081 fatcat:fwhibbpoyrdrlj5svto72rq6dm

Understanding image motion with group representations [article]

Andrew Jaegle, Stephen Phillips, Daphne Ippolito, Kostas Daniilidis
2018 arXiv   pre-print
We propose a model of motion based on elementary group properties of transformations and use it to train a representation of image motion.  ...  While most methods of estimating motion are based on pixel-level constraints, we use these group properties to constrain the abstract representation of motion itself.  ...  Here, we present a general model of visual motion and describe how the group properties of visual motion can be used to constrain learning in this model ( Figure 1 ).  ... 
arXiv:1612.00472v2 fatcat:h3zls5hmtvfqnjg67fcw75bslu

Exploiting Transformation Invariance and Equivariance for Self-supervised Sound Localisation [article]

Jinxiang Liu, Chen Ju, Weidi Xie, Ya Zhang
2022 arXiv   pre-print
of learned representations, i.e. the detected sound source should follow the same transformation applied on input video frames (transformation equivariance).  ...  We present a simple yet effective self-supervised framework for audio-visual representation learning, to localize the sound source in videos.  ...  of learned representations, e.g. audio retrieval and cross-modal retrieval.  ... 
arXiv:2206.12772v1 fatcat:4doydejbcva4be35p63dcy26ca

Geodesics of learned representations [article]

Olivier J. Hénaff, Eero P. Simoncelli
2016 arXiv   pre-print
We develop a new method for visualizing and refining the invariances of learned representations.  ...  If the transformation relating the two reference images is linearized by the representation, this sequence should follow the gradual evolution of this transformation.  ...  DISCUSSION The synthesis of geodesic sequences provide a means of visualizing and assessing metric properties of a representation.  ... 
arXiv:1511.06394v4 fatcat:j2wzlplyy5e77kjkia6gn7jore

Understand and Improve Contrastive Learning Methods for Visual Representation: A Review [article]

Ran Liu
2021 arXiv   pre-print
A promising alternative, self-supervised learning, as a type of unsupervised learning, has gained popularity because of its potential to learn effective data representations without manual labeling.  ...  Among self-supervised learning algorithms, contrastive learning has achieved state-of-the-art performance in several fields of research.  ...  Instead of predicting the representation to be representative of the property of the transformation, PIRL learns its representation to be representative of the visual semantics that are invariant of the  ... 
arXiv:2106.03259v1 fatcat:umiy7qxuinhdtjssqepl7i2xsy
« Previous Showing results 1 — 15 out of 253,878 results