No routing needed between capsules

Byerly, Adam; Kalganova, Tatiana; Dear, Ian

doi:10.1016/j.neucom.2021.08.064

Cited by 49 publications

(21 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We developed BendNet, inspired by capsule network 10,11 and homogeneous vector capsules 12 , to predict bendability of DNA duplex from sequence alone (Fig. 1a).…”

Section: Resultsmentioning

confidence: 99%

“…To extend bendability assessment to any DNA duplexes in large sequence space, we developed a deeplearning-based method termed BendNet, which extracts and learns sequence features encoding DNA bendability using capsule networks 10 without routing 11 . BendNet can predict DNA bendability of any given sequences with high agreement to that measured by loop-seq and other low throughput approaches.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Assessing base-resolution DNA mechanics on the genome scale

Jiang

et al. 2022

Preprint

View full text Add to dashboard Cite

Intrinsic DNA properties such as bending play a crucial role in diverse biological systems. Recent advantage in highthroughput method called loop-seq makes it possible to determine bendability of hundred thousand 50bp DNA duplexes in one experiment. However, it is still infeasible to assess whole sequence bendability in large genomes such as human, which needs thousands loop-seq experiments. Here we introduce BendNet: a neural network to accurately predict the intrinsic DNA bending at base-resolution by only given DNA sequences. Bendnet can increase the resolution of experimental results, and can predict DNA bendability for any new given sequences in high accuracy. We applied bendnet to the human genome and observed high-stiffness regions located at both transcriptional start sites and transcriptional end sites. Such stiffness patterns are different for coding and non-coding genes, which matches distinct nucleosome occupancy patterns. As expected, most of transcription factors (TFs) bind in DNA of low bendability. In contrast, we observed an unusually high bendability within binding elements of specific TFs such as EBF1 and regulators of genome folding such as CTCF. These factors either co-bind or compete with nucleosome to carry out their functions. More interestingly, CTCF binding regions exhibit the highest bendability than other DNA elements, implying their potential role in trapping and holding the CTCF in the exact locations to make sure CTCF as stable anchor in loop extrusion process. Our work provides a tool to assess DNA bendability for large scale DNA sequences and expands our understanding on DNA mechanics in chromatin regulation and genome folding.

show abstract

“…We developed BendNet, inspired by capsule network 10,11 and homogeneous vector capsules 12 , to predict bendability of DNA duplex from sequence alone (Fig. 1a).…”

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Assessing base-resolution DNA mechanics on the genome scale

Jiang

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The naming of network architectures follows 75 , where layers are separated by—and spatial dimensions are separated by ×. The convolution layer and pooling layer are represented by C and P, and the hyperdimension memory, unique to , is represented by H. Model and algorithms Network architecture Accuracy (%) DECOLLE 16 16C7-P2-32 32C7-P2-64 98.0 EventProp 78 784-350-10 97.6 STDBP 75 28 28-16C5-P2-32C5-P2-800-128-10 99.4 HVCs 88 *State-of-the-art ANN model 99.8 SpikeHD 16 16C7-P2-32 32C7-P2-64-4kH 99.2 …”

Section: Discussionmentioning

confidence: 99%

Memory-inspired spiking hyperdimensional network for robust online learning

Zou

Alimohamadi

Zakeri

et al. 2022

Sci Rep

View full text Add to dashboard Cite

Recently, brain-inspired computing models have shown great potential to outperform today’s deep learning solutions in terms of robustness and energy efficiency. Particularly, Spiking Neural Networks (SNNs) and HyperDimensional Computing (HDC) have shown promising results in enabling efficient and robust cognitive learning. Despite the success, these two brain-inspired models have different strengths. While SNN mimics the physical properties of the human brain, HDC models the brain on a more abstract and functional level. Their design philosophies demonstrate complementary patterns that motivate their combination. With the help of the classical psychological model on memory, we propose SpikeHD, the first framework that fundamentally combines Spiking neural network and hyperdimensional computing. SpikeHD generates a scalable and strong cognitive learning system that better mimics brain functionality. SpikeHD exploits spiking neural networks to extract low-level features by preserving the spatial and temporal correlation of raw event-based spike data. Then, it utilizes HDC to operate over SNN output by mapping the signal into high-dimensional space, learning the abstract information, and classifying the data. Our extensive evaluation on a set of benchmark classification problems shows that SpikeHD provides the following benefit compared to SNN architecture: (1) significantly enhance learning capability by exploiting two-stage information processing, (2) enables substantial robustness to noise and failure, and (3) reduces the network size and required parameters to learn complex information.

show abstract

“…When training this network, a batch size of 120 was used. Data augmentation used during the training for these datasets was uniform and the same strategy as described in [35].…”

Section: Training On Mnist and Fashion-mnistmentioning

confidence: 99%

Towards an Analytical Definition of Sufficient Data

Byerly¹,

Kalganova²

2022

Preprint

Self Cite

View full text Add to dashboard Cite

We show that, for each of five datasets of increasing complexity, certain training samples are more informative of class membership than others. These samples can be identified a priori to training by analyzing their position in reduced dimensional space relative to the classes' centroids. Specifically, we demonstrate that samples nearer the classes' centroids are less informative than those that are furthest from it. For all five datasets, we show that there is no statistically significant difference between training on the entire training set and when excluding up to 2% of the data nearest to each class's centroid.

show abstract

No routing needed between capsules

Cited by 49 publications

References 5 publications

Assessing base-resolution DNA mechanics on the genome scale

Assessing base-resolution DNA mechanics on the genome scale

Memory-inspired spiking hyperdimensional network for robust online learning

Towards an Analytical Definition of Sufficient Data

Contact Info

Product

Resources

About