More Like this

1. Image classification and training with severe data loss

   https://doi.org/10.1117/12.2633172  

   Marquard, Dillon; Wright, Kyle; Marcia, Roummel F. (October 2022, SPIE Optical Engineering + Applications, 2022)

   Zelinski, Michael E.; Taha, Tarek M.; Howe, Jonathan (Ed.)

   Image classification forms an important class of problems in machine learning and is widely used in many realworld applications, such as medicine, ecology, astronomy, and defense. Convolutional neural networks (CNNs) are machine learning techniques designed for inputs with grid structures, e.g., images, whose features are spatially correlated. As such, CNNs have been demonstrated to be highly effective approaches for many image classification problems and have consistently outperformed other approaches in many image classification and object detection competitions. A particular challenge involved in using machine learning for classifying images is measurement data loss in the form of missing pixels, which occurs in settings where scene occlusions are present or where the photodetectors in the imaging system are partially damaged. In such cases, the performance of CNN models tends to deteriorate or becomes unreliable even when the perturbations to the input image are small. In this work, we investigate techniques for improving the performance of CNN models for image classification with missing data. In particular, we explore training on a variety of data alterations that mimic data loss for producing more robust classifiers. By optimizing the categorical cross-entropy loss function, we demonstrate through numerical experiments on the MNIST dataset that training with these synthetic alterations can enhance the classification accuracy of our CNN models. 

   more » « less
2. Signature Driven Post-Manufacture Testing and Tuning of RRAM Spiking Neural Networks for Yield Recovery

   Saha, A; Amarnath, C; Ma, K; Chatterjee, A (May 2025, Proceeding, IEEE/SIGDA Asian South Pacific Design Automation Conference)

   IEEE (Ed.)

   Resistive random access Memory (RRAM) based spiking neural networks (SNN) are becoming increasingly attractive for pervasive energy-efficient classification tasks. However, such networks suffer from degradation of performance (as determined by classification accuracy) due to the effects of process variations on fabricated RRAM devices resulting in loss of manufacturing yield. To address such yield loss, a two-step approach is developed. First, an alternative test framework is used to predict the performance of fabricated RRAM based SNNs using the SNN response to a small subset of images from the test image dataset, called the SNN response signature (to minimize test cost). This diagnoses those SNNs that need to be performance-tuned for yield recovery. Next, SNN tuning is performed by modulating the spiking thresholds of the SNN neurons on a layer-by-layer basis using a trained regressor that maps the SNN response signature to the optimal spiking thresholdvalues during tuning. The optimal spiking threshold values are determined by an off-line optimization algorithm. Experiments show that the proposed framework can reduce the number of out-of-spec SNN devices by up to 54% and improve yield by as much as 8.6%. 

   more » « less
3. Intelligent nanoscope for rapid nanomaterial identification and classification

   https://doi.org/10.1039/D2LC00206J  

   Jin, Geonsoo; Hong, Seongwoo; Rich, Joseph; Xia, Jianping; Kim, Kyeri; You, Lingchong; Zhao, Chenglong; Huang, Tony Jun (June 2022, Lab on a Chip)

   Machine learning image recognition and classification of particles and materials is a rapidly expanding field. However, nanomaterial identification and classification are dependent on the image resolution, the image field of view, and the processing time. Optical microscopes are one of the most widely utilized technologies in laboratories across the world, due to their nondestructive abilities to identify and classify critical micro-sized objects and processes, but identifying and classifying critical nano-sized objects and processes with a conventional microscope are outside of its capabilities, due to the diffraction limit of the optics and small field of view. To overcome these challenges of nanomaterial identification and classification, we developed an intelligent nanoscope that combines machine learning and microsphere array-based imaging to: (1) surpass the diffraction limit of the microscope objective with microsphere imaging to provide high-resolution images; (2) provide large field-of-view imaging without the sacrifice of resolution by utilizing a microsphere array; and (3) rapidly classify nanomaterials using a deep convolution neural network. The intelligent nanoscope delivers more than 46 magnified images from a single image frame so that we collected more than 1000 images within 2 seconds. Moreover, the intelligent nanoscope achieves a 95% nanomaterial classification accuracy using 1000 images of training sets, which is 45% more accurate than without the microsphere array. The intelligent nanoscope also achieves a 92% bacteria classification accuracy using 50 000 images of training sets, which is 35% more accurate than without the microsphere array. This platform accomplished rapid, accurate detection and classification of nanomaterials with miniscule size differences. The capabilities of this device wield the potential to further detect and classify smaller biological nanomaterial, such as viruses or extracellular vesicles. 

   more » « less
4. On Representation Learning for Road Networks

   https://doi.org/10.1145/3424346  

   Wang, Meng-Xiang; Lee, Wang-Chien; Fu, Tao-Yang; Yu, Ge (February 2021, ACM Transactions on Intelligent Systems and Technology)

   Informative representation of road networks is essential to a wide variety of applications on intelligent transportation systems. In this article, we design a new learning framework, called Representation Learning for Road Networks (RLRN), which explores various intrinsic properties of road networks to learn embeddings of intersections and road segments in road networks. To implement the RLRN framework, we propose a new neural network model, namely Road Network to Vector (RN2Vec), to learn embeddings of intersections and road segments jointly by exploring geo-locality and homogeneity of them, topological structure of the road networks, and moving behaviors of road users. In addition to model design, issues involving data preparation for model training are examined. We evaluate the learned embeddings via extensive experiments on several real-world datasets using different downstream test cases, including node/edge classification and travel time estimation. Experimental results show that the proposed RN2Vec robustly outperforms existing methods, including (i) Feature-based methods : raw features and principal components analysis (PCA); (ii) Network embedding methods : DeepWalk, LINE, and Node2vec; and (iii) Features + Network structure-based methods : network embeddings and PCA, graph convolutional networks, and graph attention networks. RN2Vec significantly outperforms all of them in terms of F1-score in classifying traffic signals (11.96% to 16.86%) and crossings (11.36% to 16.67%) on intersections and in classifying avenue (10.56% to 15.43%) and street (11.54% to 16.07%) on road segments, as well as in terms of Mean Absolute Error in travel time estimation (17.01% to 23.58%). 

   more » « less
5. ReSpike: residual frames-based hybrid spiking neural networks for efficient action recognition

   https://doi.org/10.1088/2634-4386/adb070  

   Xiao, Shiting; Li, Yuhang; Kim, Youngeun; Lee, Donghyun; Panda, Priyadarshini (January 2025, Neuromorphic Computing and Engineering)

   Abstract Spiking Neural Networks (SNNs) have emerged as a compelling, energy-efficient alternative to traditional Artificial Neural Networks (ANNs) for static image tasks such as image classification and segmentation. However, in the more complex video classification domain, SNN-based methods fall considerably short of ANN-based benchmarks, due to the challenges in processing dense RGB frames. To bridge this gap, we propose ReSpike, a hybrid framework that synergizes the strengths of ANNs and SNNs to tackle action recognition tasks with high accuracy and low energy cost. By partitioning film clips into RGB image Key Frames, which primarily capture spatial information, and event-like Residual Frames, which emphasize temporal dynamics cues, ReSpike leverages ANN for processing spatial features and SNN for modeling temporal features. In addition, we propose a multi-scale cross-attention mechanism for effective feature fusion. Compared to state-of-the-art SNN baselines, our ReSpike hybrid architecture demonstrates significant performance improvements (e.g., >30% absolute accuracy improvement on both HMDB-51 and UCF-101 datasets). Additionally, ReSpike is the first SNN method capable of scaling to the large-scale benchmark Kinetics-400. Furthermore, ReSpike achieves comparable performance with prior ANN approaches while bringing better accuracy-energy tradeoff. 

   more » « less