An official website of the United States government
Spiking Neural Networks (SNNs) have recently become more popular as a biologically plausible substitute for traditional Artificial Neural Networks (ANNs). SNNs are cost-efficient and deployment-friendly because they process input in both spatial and temporal manner using binary spikes. However, we observe that the information capacity in SNNs is affected by the number of timesteps, leading to an accuracy-efficiency tradeoff. In this work, we study a fine-grained adjustment of the number of timesteps in SNNs. Specifically, we treat the number of timesteps as a variable conditioned on different input samples to reduce redundant timesteps for certain data. We call our method Spiking Early-Exit Neural Networks (SEENNs). To determine the appropriate number of timesteps, we propose SEENN-I which uses a confidence score thresholding to filter out the uncertain predictions, and SEENN-II which determines the number of timesteps by reinforcement learning. Moreover, we demonstrate that SEENN is compatible with both the directly trained SNN and the ANN-SNN conversion. By dynamically adjusting the number of timesteps, our SEENN achieves a remarkable reduction in the average number of timesteps during inference. For example, our SEENN-II ResNet-19 can achieve 96.1% accuracy with an average of 1.08 timesteps on the CIFAR-10 test dataset. Code is shared at https://github.com/Intelligent-Computing-Lab-Yale/SEENN.
more »
« less
Robust Bio-Secure DNA Based Memory
Information storage in synthetic DNA oligomers is attractive due to the inherent physical density, stability, and energy efficiency of nucleic acids. Information retention –during writing, storage, and retrieval processes– requires development of efficient encoding/decoding systems. Additionally, potential intrusion of artificial or organic malevolent biologically active molecular machines could potentially cause catastrophic biosecurity concerns. Here we present an improved information storage method that focuses on efficiency and biosecurity. Herein this paper, we have developed and experimentally tested an algorithm to write data in pool of DNA strands by applying a fountain code (rateless erasure code), a Reed Solomon code, and an oligomer mapping code that ensures Bio-Security. We validated our method through wet-lab experiments and wrote, stored, and fully retrieved 105,360 bits of information. We validated the biosecurity aspects of our method through in-silico experimentation using a BLAST-run to compare our generated oligomers to existing genes documented in the public databases, a Plasmidhawk software analysis to determine our oligomers could not be artificially traced to have originated from another lab, and utilized an open-source software to determine whether our oligomers could have expressed any sequences that potentially originate or empower biologically meaningful functions.
more »
« less
- Award ID(s):
- 2027738
- PAR ID:
- 10509344
- Publisher / Repository:
- Duke
- Date Published:
- Journal Name:
- Foundations of Nanoscience (FNANO)
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Pathogenic dsDNA prompts AIM2 assembly leading to the formation of the inflammasome, a multimeric complex that triggers the inflammatory response. The recognition of foreign dsDNA involves AIM2 self-assembly concomitant with dsDNA binding. However, we lack mechanistic and kinetic information on the formation and propagation of the assembly, which can shed light on innate immunity’s time response and specificity. Combining optical traps and confocal fluorescence microscopy, we determine here the association and dissociation rates of the AIM2-DNA complex at the single molecule level. We identify distinct mechanisms for oligomer growth via the binding of incoming AIM2 molecules to adjacent dsDNA or direct interaction with bound AIM2 assemblies, resembling primary and secondary nucleation. Through these mechanisms, the size of AIM2 oligomers can increase fourfold in seconds. Finally, our data indicate that single AIM2 molecules do not diffuse/scan along the DNA, suggesting that oligomerization depends on stochastic encounters with DNA and/or DNA-bound AIM2.more » « less
-
Abstract DNA is a compelling alternative to non-volatile information storage technologies due to its information density, stability, and energy efficiency. Previous studies have used artificially synthesized DNA to store data and automated next-generation sequencing to read it back. Here, we report digital Nucleic Acid Memory (dNAM) for applications that require a limited amount of data to have high information density, redundancy, and copy number. In dNAM, data is encoded by selecting combinations of single-stranded DNA with (1) or without (0) docking-site domains. When self-assembled with scaffold DNA, staple strands form DNA origami breadboards. Information encoded into the breadboards is read by monitoring the binding of fluorescent imager probes using DNA-PAINT super-resolution microscopy. To enhance data retention, a multi-layer error correction scheme that combines fountain and bi-level parity codes is used. As a prototype, fifteen origami encoded with ‘Data is in our DNA!\n’ are analyzed. Each origami encodes unique data-droplet, index, orientation, and error-correction information. The error-correction algorithms fully recover the message when individual docking sites, or entire origami, are missing. Unlike other approaches to DNA-based data storage, reading dNAM does not require sequencing. As such, it offers an additional path to explore the advantages and disadvantages of DNA as an emerging memory material.more » « less
-
One of the most significant challenges in the field of software code auditing is the presence of vulnerabilities in software source code. Every year, more and more software flaws are discovered, either internally in proprietary code or publicly disclosed. These flaws are highly likely to be exploited and can lead to system compromise, data leakage, or denial of service. To create a large-scale machine learning system for function-level vulnerability identification, we utilized a sizable dataset of C and C++ open-source code containing millions of functions with potential buffer overflow exploits. We have developed an efficient and scalable vulnerability detection method based on neural network models that learn features extracted from the source codes. The source code is first converted into an intermediate representation to remove unnecessary components and shorten dependencies. We maintain the semantic and syntactic information using state-ofthe- art word embedding algorithms such as GloVe and fastText. The embedded vectors are subsequently fed into neural networks such as LSTM, BiLSTM, LSTM-Autoencoder, word2vec, BERT, and GPT-2 to classify the possible vulnerabilities. Furthermore, we have proposed a neural network model that can overcome issues associated with traditional neural networks. We have used evaluation metrics such as F1 score, precision, recall, accuracy, and total execution time to measure the performance. We have conducted a comparative analysis between results derived from features containing a minimal text representation and semantic and syntactic information. We have found that all neural network models provide higher accuracy when we use semantic and syntactic information as features. However, this approach requires more execution time due to the added complexity of the word embedding algorithm. Moreover, our proposed model provides higher accuracy than LSTM, BiLSTM, LSTM-Autoencoder, word2vec and BERT models, and the same accuracy as the GPT-2 model with greater efficiency.more » « less
-
James Aspnes and Othon Michail (Ed.)SIMD||DNA [Wang et al., 2019] is a model of DNA strand displacement allowing parallel in-memory computation on DNA storage. We show how to simulate an arbitrary 3-symbol space-bounded Turing machine with a SIMD||DNA program, giving a more direct and efficient route to general-purpose information manipulation on DNA storage than the Rule 110 simulation of Wang, Chalk, and Soloveichik [Wang et al., 2019]. We also develop software (https://github.com/UC-Davis-molecular-computing/simd-dna) that can simulate SIMD||DNA programs and produce SVG figures.more » « less
