{"total":15,"items":[{"citing_arxiv_id":"2606.18122","ref_index":4,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Embedded Machine Learning for Microcontroller-Class Edge Devices: Data, Feature, Evaluation, and Deployment Pipelines","primary_cat":"cs.LG","submitted_at":"2026-06-16T16:22:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":2.0,"formal_verification":"none","one_line_summary":"This paper outlines a systems-oriented workflow for embedded machine learning on microcontrollers, using accelerometer-based motion recognition and audio keyword spotting as running examples to illustrate data, feature, evaluation, and deployment steps.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.17249","ref_index":7,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"From Compression to Deployment: Real-Time and Energy-Efficient FastGRNN on Ultra-Constrained Microcontrollers","primary_cat":"cs.AR","submitted_at":"2026-06-15T19:49:38+00:00","verdict":"ACCEPT","verdict_confidence":"MODERATE","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Compressed FastGRNN model with 566-byte weights runs real-time 50 Hz inference on Arduino and MSP430 MCUs at macro F1 0.918 while matching PyTorch reference and cutting energy 96.7% via LUT.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.07470","ref_index":49,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Verifiable and Confidential DNN Inference on Low-End Edge Devices","primary_cat":"cs.CR","submitted_at":"2026-06-05T17:23:39+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"VECODI introduces SHANGRI-LA, an intermediate-privilege runtime on TrustZone-M, to enable verifiable confidential DNN inference on constrained edge devices with small TCB and overhead.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.06528","ref_index":12,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Quantized AI Inference on Constrained Embedded Platforms for Small-Satellite Settings","primary_cat":"cs.AR","submitted_at":"2026-06-03T10:31:13+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":3.0,"formal_verification":"none","one_line_summary":"Measurement-based characterization of quantized AI inference latency and data movement on Cortex-M platforms, positioned as a lower-bound reference for small-satellite embedded vision workloads.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.20853","ref_index":78,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"SEABAD: A Tropical Bird Activity Detection Dataset for Passive Acoustic Monitoring","primary_cat":"cs.SD","submitted_at":"2026-05-20T07:44:39+00:00","verdict":"ACCEPT","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"SEABAD is a publicly released, balanced dataset of 50,000 curated 16 kHz audio clips spanning 1,677 tropical bird species, with a dual-branch curation pipeline and MobileNetV3-Small baseline reaching 99.57% accuracy.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.18704","ref_index":62,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Learned Memory Attenuation in Sage-Husa Kalman Filters for Robust UAV State Estimation","primary_cat":"eess.SP","submitted_at":"2026-05-18T17:38:13+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"NDR-SHKF replaces the static forgetting factor in Sage-Husa Kalman Filters with a learned vector-valued memory attenuation policy from a bifurcated recurrent network trained end-to-end on whitened innovations to minimize estimation error.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.15694","ref_index":24,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Going Beyond the Edge: Distributed Inference of Transformer Models on Ultra-Low-Power Wireless Devices","primary_cat":"cs.LG","submitted_at":"2026-05-15T07:33:53+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"CATS enables collaborative transformer inference on up to 16 ultra-low-power wireless devices, supporting models up to 14 times larger than a single device can run via SomeGather pruning and message-dropout robustness.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.09357","ref_index":27,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Split CNN Inference on Networked Microcontrollers","primary_cat":"cs.DC","submitted_at":"2026-05-10T06:16:11+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A fine-grained split inference system enables CNN models infeasible on single MCUs to run across networked devices by partitioning at sub-layer granularity, reducing per-device peak RAM while keeping practical latency.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"redundant connections or neurons to accelerate inference. Beyond static compression, hardware-aware neural architec- ture search, such as MCUNet [10], automates the design of efficient network topologies tailored for the strict constraints of MCUs. At the system level, lightweight inference frame- works such as TensorFlow Lite for Microcontrollers [26] and CMSIS-NN [27] optimize kernel implementations to minimize latency. However, although they offer a valuable way to optimize the inference, the deployment of large-scale DNNs remains bounded by the physical limitations of a single MCU. This bottleneck necessitates a shift from single-device optimization to a distributed inference strategy. Distributed Inference for Edge Devices."},{"citing_arxiv_id":"2605.03039","ref_index":88,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Mixed-Precision Information Bottlenecks for On-Device Trait-State Disentanglement in Bipolar Agitation Detection","primary_cat":"cs.LG","submitted_at":"2026-05-04T18:06:17+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"MP-IB uses an 8x information asymmetry via FP16 trait heads and INT4 state heads to disentangle speaker identity from agitation in voice biomarkers, outperforming larger models on edge devices with low latency and suppressed identity leakage.","context_count":1,"top_context_role":"baseline","top_context_polarity":"baseline","context_text":"Enrollment: 3 utterances per speaker; testing: 10 utterances per speaker; 1,200 total trials; 95% CIs via bootstrap. Method Top-1↓Top-5↓EER↓MI (bits)↓MIA-AUC↓ Random Features (128-d) 0.009 [0.007,0.011] 0.042 [0.038,0.046] 0.48 [0.46,0.50] 0.12 [0.10,0.14] 0.51 [0.48,0.54] Uniform FP16 SER 0.45 [0.42,0.48] 0.72 [0.69,0.75] 0.12 [0.10,0.14] 4.8 [4.5,5.1] 0.85 [0.82,0.88] Adversarial-MLP 0.38 [0.35,0.41] 0.65 [0.62,0.68] 0.18 [0.16,0.20] 3.9 [3.6,4.2] 0.78 [0.75,0.81] ECAPA-TDNN-Adapter 0.21 [0.19,0.23] 0.48 [0.45,0.51] 0.22 [0.20,0.24] 2.7 [2.4,3.0] 0.71 [0.68,0.74] WavLM-Adapter 0.33 [0.30,0.36] 0.61 [0.58,0.64] 0.15 [0.13,0.17] 4.1 [3.8,4.4] 0.82 [0.79,0.85] β-V AE Disentanglement 0.28 [0.25,0.31] 0.52 [0.49,0.55] 0."},{"citing_arxiv_id":"2604.27004","ref_index":45,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"EdgeSpike: Spiking Neural Networks for Low-Power Autonomous Sensing in Edge IoT Architectures","primary_cat":"cs.NE","submitted_at":"2026-04-29T05:15:28+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"EdgeSpike delivers 91.4% mean accuracy on five sensing tasks with 31x lower energy on neuromorphic hardware and 6.3x longer battery life in a seven-month field deployment compared to conventional CNNs.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.16113","ref_index":7,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Co-Design of CNN Accelerators for TinyML using Approximate Matrix Decomposition","primary_cat":"cs.AR","submitted_at":"2026-04-17T14:49:17+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A co-design framework using approximate matrix decomposition and genetic algorithms delivers 33% average latency reduction in TinyML CNN FPGA accelerators with 1.3% average accuracy loss versus standard systolic arrays.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Conference on Computer Vision and Pattern Recognition (CVPR), (Las Vegas, NV , USA), IEEE, 2016, pp. 779-788.DOI: 10.1109/CVPR.2016. 91 [6] STMicroelectronics,X-CUBE-AI - artificial intelligence (AI) soft- ware expansion for STM32CubeMX, https://www.st.com/en/embedded- software/x-cube-ai.html, version DB3788 - Rev 11, 2024. Accessed: Mar. 28, 2026. [7] L. Lai, N. Suda, and V . Chandra, \"CMSIS-NN: Efficient neural network kernels for Arm Cortex-M CPUs,\"The Computing Research Repository (CoRR), 2018. arXiv: 1801.06601[cs.NE]. [8] T. Chen et al., \"TVM: An automated end-to-end optimizing compiler for deep learning,\"The Computing Research Repository (CoRR), 2018. arXiv: 1802.04799[cs.LG]. [9] J. Lin, W."},{"citing_arxiv_id":"2604.15714","ref_index":11,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Neuromorphic Parameter Estimation for Power Converter Health Monitoring Using Spiking Neural Networks","primary_cat":"cs.NE","submitted_at":"2026-04-17T05:34:53+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A three-layer leaky integrate-and-fire spiking neural network estimates passive component parameters in power converters, cutting resistance error from 25.8% to 10.2% versus feedforward baselines at projected 270x lower energy on neuromorphic chips.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.09565","ref_index":20,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"AEG: A Baremetal Framework for AI Acceleration via Direct Hardware Access in Heterogeneous Accelerators","primary_cat":"cs.DC","submitted_at":"2026-02-15T22:12:45+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"AEG baremetal framework achieves 9.2x higher compute efficiency, 3-7x less data movement, and near-zero latency variance for ResNet-18 on 28 AIE tiles versus Linux Vitis AI on 304 tiles while maintaining 68.78% ImageNet accuracy.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2602.06523","ref_index":32,"ref_count":2,"confidence":0.98,"is_internal_anchor":true,"paper_title":"MicroBi-ConvLSTM: An Ultra-Lightweight Efficient Model for Human Activity Recognition on Resource Constrained Devices","primary_cat":"cs.CV","submitted_at":"2026-02-06T09:26:29+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"MicroBi-ConvLSTM is a convolutional-recurrent model with 11.4K parameters that delivers competitive accuracy on eight HAR benchmarks and full INT8 deployment coverage on Raspberry Pi Pico 2 and ESP32.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"1806.00582","ref_index":2,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Federated Learning with Non-IID Data","primary_cat":"cs.LG","submitted_at":"2018-06-02T04:45:58+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Non-IID data causes up to 55% accuracy loss in federated learning due to weight divergence measured by earth mover's distance; 5% globally shared data recovers 30% accuracy on CIFAR-10.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}