Pith. sign in

REVIEW 4 major objections 5 minor 41 references

UPAQ: A Framework for Real-Time and Energy-Efficient 3D Object Detection in Autonomous Vehicles

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read UPAQ claims to compress 3D object detectors up to 5.62× while making them faster, more energy-efficient, and more accurate on embedded hardware.

desk verdict UPAQ combines known pruning and quantization ideas in a practically oriented way, but the missing definition of its efficiency score and the unverifiable mAP gains make the central claims unsupported as written. read the letter →

arxiv 2501.04213 v1 pith:HA6LIMR3 submitted 2025-01-08 cs.CV cs.AIcs.LG

Abhishek Balasubramaniam , Febin P Sunny , Sudeep Pasricha This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords 3Dobjectdetectionpatternpruningmixed-precisionquantizationmodelcompressionembeddedplatformsautonomousvehiclesPointPillarsSMOKE
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

UPAQ is a compression framework for 3D object detectors in autonomous vehicles, combining pattern-based pruning (keeping only a few weights per kernel in diagonal, row, or column layouts) with mixed-precision quantization (different bit widths for different weights). The paper claims that on the Jetson Orin Nano embedded platform, UPAQ compresses the PointPillars LiDAR detector 5.62× and the SMOKE camera detector 5.13×, while making inference up to 1.97× and 1.86× faster and cutting energy use up to 2.07× and 1.87× compared with state-of-the-art pruning and quantization methods. In its accuracy-focused configuration, UPAQ also raises mean average precision without any retraining, from 78.96 to 86.15 on PointPillars and from 29.85 to 36.65 on SMOKE. If these numbers hold, compressed 3D perception could run in real time on the low-power hardware typically found in vehicles, addressing a major practical bottleneck for autonomous driving.

What carries the argument

The framework's core selection device is the efficiency score $E_s = \alpha\cdot\mathrm{SQNR} + \beta\cdot(1/\mathrm{Latency}) + \gamma\cdot(1/\mathrm{Energy})$, which scores each candidate pruned-and-quantized kernel by combining its signal-to-quantization-noise ratio with the reciprocal of on-device latency and energy. For $k\times k$ kernels with $k>1$, the method generates candidate sparsity patterns (main diagonal, anti-diagonal, row, or column placements of non-zero weights), quantizes the surviving weights with a symmetric mixed-precision quantizer, and keeps the pattern with the highest $E_s$; for $1\times 1$ kernels, it flattens and regroups them into $k\times k$ blocks so the same pattern machinery applies. A preprocessing stage groups layers into root-leaf sets by depth-first search, so the expensive search runs only on root layers and the winning pattern is replicated on coupled leaf layers. The two reported configurations, HCK (high compression) and LCK (low compression), correspond to fewer non-zero weights per kernel plus lower bit widths (e.g., 4 and 8 bits) versus more non-zero weights plus higher bit widths (e.g., 8 and 16 bits).

What would settle it

Measure the per-kernel latency and energy of every candidate pruned-and-quantized kernel on the actual Jetson Orin Nano, recompute the efficiency score with those measured values, and verify that the kernels UPAQ selects still produce the reported 1.97× and 1.86× speedups and 2.07× and 1.87× energy reductions; if the top-scoring kernels change and the gains disappear, the central claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that a two-tier compression scheme—$k\times k$ kernel compression with a pattern generator (main-diagonal, anti-diagonal, row, or column placements of non-zero weights) and a $1\times 1$-to-$k\times k$ reshape trick for point-cloud feature layers—can simultaneously shrink, speed up, and improve the accuracy of 3D object detectors. UPAQ chooses the best compression per root layer via an efficiency score that balances quantization noise against latency and energy, then propagates that pattern to coupled leaf layers. The paper reports 5.62× compression for PointPillars and 5.13× for SMOKE, up to 1.97× and 1.86× inference speedups on the Jetson Orin, up to 2.07× and 1.87× energy reduction, and higher mAP than the uncompressed baseline in the accuracy-oriented LCK variant, all with no retraining. The framework is presented as beating all prior pruning and quantization methods on every reported axis.

Load-bearing premise

The load-bearing premise is that the efficiency score's latency and energy terms come from a reliable model of on-device performance; the paper never specifies how those values are measured or predicted, so if that model is wrong, the reported speedups and energy savings may be an artifact of the scoring function rather than the compression.

Editorial extensions

If this is right

  • If the reported numbers hold, PointPillars inference drops to 18.23 ms and SMOKE to 68.45 ms on the Jetson Orin, putting the LiDAR detector inside a real-time budget and nearly halving the camera detector's latency.
  • The LCK variant raises mean average precision from 78.96 to 86.15 on PointPillars and from 29.85 to 36.65 on SMOKE without retraining, showing that pruning and quantization need not cost accuracy.
  • Because the expensive search runs only on root layers and the chosen pattern is copied to leaf layers, compression cost scales with the number of root layers rather than the total layer count, which matters for deeper detectors.
  • Both a LiDAR-based detector and a camera-based detector improve, so the method is not tied to one input modality among 3D perception pipelines.
  • The HCK configuration gives users a compression-first option and the LCK configuration an accuracy-first option, a tunable trade-off controlled by the number of non-zero weights per kernel and the quantization bit widths.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not test per-layer tuning of the efficiency-score weights (it fixes α=0.3, β=0.4, γ=0.3), so a natural extension is to search those weights per layer to see if the reported latency-energy trade-off can be pushed further.
  • The $1\times 1$-to-$k\times k$ reshape specifically targets point-cloud feature networks, so isolating the Pillar Feature Network in an ablation would reveal how much of the gain comes from that trick versus the root-leaf search.
  • The related work lists larger 3D detectors (SECOND, Focals Conv, VSC); applying UPAQ to those would test whether the compression and speedup numbers scale beyond the two architectures studied.
  • Since LCK improves accuracy without retraining, a short fine-tuning pass after compression might compound the gain, but the paper does not explore that combination.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes UPAQ, a two-stage compression framework for 3D object detectors that combines semi-structured pattern pruning with mixed-precision quantization. It is evaluated on PointPillars and SMOKE, targeting embedded platforms such as the Jetson Orin Nano. The authors report high compression ratios (up to 5.62x), inference speedups (up to 1.97x), energy reductions (up to 2.07x), and mAP improvements over the uncompressed baselines and over four prior compression methods. The method is centered on a per-kernel efficiency score Es (Eq. 2) that is used to select pruning and quantization choices in Algorithms 4 and 5.

Significance. If the reported results are correct, UPAQ would be a significant advance for real-time embedded 3D detection, since it claims to simultaneously improve compression, latency, energy, and accuracy over strong baselines, including an mAP gain of about 7 points without retraining. The paper also targets 1x1 kernel compression, which is often neglected. However, the central selection mechanism is underspecified and the evaluation protocol is incomplete, so the validity of the headline claims cannot be assessed from the manuscript as written. These gaps are load-bearing for the paper's contributions and must be addressed before the work can be considered reproducible.

major comments (4)
  1. [Section IV.C.3, Eq. (2), Algorithms 4 and 5] The efficiency score Es is the sole criterion for selecting every pruned and quantized kernel, but the manuscript never specifies how the on-device latency and energy of a candidate compressed model are computed. Section IV.C.3 states only that 'we then calculate the on-device latency and energy of the model' and insert these values into Eq. (2); no surrogate model, measurement procedure, calibration, or validation is given. Because Es directly determines every kernel selection in Algorithms 4 and 5, the reported speedups and energy reductions in Table 2 cannot be attributed to the described method. An independent implementation would have to guess the latency and energy model, making the algorithm irreproducible in its current form.
  2. [Section V.A, Table 2] The evaluation protocol is not reported. The paper states a KITTI 80:10:10 split but does not specify whether this is the standard KITTI train/validation split, which object class and difficulty level are used, what IoU threshold is applied, what mAP variant is reported (e.g., 3D AP, BEV AP, R11/R40), or whether results are averaged over multiple runs. These details are essential for comparing against the cited baselines and for assessing the accuracy claims. Without them, the mAP values in Table 2 (e.g., 86.15 for PointPillars and 36.65 for SMOKE under UPAQ-LCK) cannot be verified or reproduced.
  3. [Section V.B, Table 2] The reported mAP improvements without retraining are anomalous and unexplained. UPAQ (LCK) improves PointPillars mAP from 78.96 to 86.15 and SMOKE from 29.85 to 36.65, while applying pruning and quantization to an already converged model with no fine-tuning. This goes against the typical behavior of post-training quantization and pruning, which generally cause accuracy loss or at most minor changes. The paper offers no mechanism for this improvement and no additional analysis, such as per-class AP, error analysis, or repeated evaluations, to rule out an evaluation artifact. This claim needs to be substantiated before the paper's dominance over prior methods can be accepted.
  4. [Algorithms 4 and 5] The pseudocode is not executable as printed and contains undefined operations. Algorithm 4 line 6 uses temp_array before it is defined (line 3 creates temp_kernel); Algorithm 5 line 11 appends to temp_array without initialization, line 12 uses the invalid expression t1=0, and line 19 calls Flatten on a list; Algorithm 2 uses ranges written as '(0, min(n, d)]', which is not valid Python. In addition, Algorithm 4 mutates KW inside the pattern loop (line 10) without resetting temp_kernel or restoring the original weights, so each candidate is not evaluated independently and the selection is order-dependent. These issues prevent reproduction of the method and cast doubt on the correctness of the described search procedure.
minor comments (5)
  1. [Table 2] The term 'Compression' is not defined; please state the exact formula (e.g., original bit count divided by compressed bit count) and report per-model sparsity and bitwidth distributions so that the compression ratios can be interpreted.
  2. [Section V.A] The paper does not specify the exact pretrained checkpoints or training configurations for PointPillars and SMOKE; please provide model sources and any required preprocessing steps.
  3. [References and Section V.B] Reference numbering is inconsistent: Ps&Qs is cited as [20] in the related work but [21] in Section V.B, and LIDAR-PTQ is cited as [22] in the related work and [23] later; please harmonize all citations.
  4. [Abstract and Section V.A] The abstract says 'Jetson Orin Nano' while Section V.A says 'Jetson Orin'; please use the exact platform name consistently throughout.
  5. [Section IV.B, Algorithm 2] The range notation such as '(0, min(n, d)]' is ambiguous; use Python range() or standard mathematical interval notation consistently.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: UPAQ's reported gains are measured outcomes of a stated optimization objective, not derived from that objective by construction.

full rationale

The reported compression, mAP, latency, and energy numbers in Table 2 are measured on the RTX 4080 and Jetson Orin platforms, not computed from the efficiency score Es. The efficiency score in Eq. (2) is an optimization objective used to select among candidate pruned and quantized kernels; including 1/Latency and 1/Energy in that objective and then measuring the same metrics after deployment is direct optimization, not circular prediction. The mAP improvements (78.96 to 86.15 for PointPillars, 29.85 to 36.65 for SMOKE) are not part of Es and provide independent evidence that the compression choices are not merely fitting the reported metrics. The only substantive concern is that Section IV.C does not specify how per-kernel on-device latency and energy are obtained before Es is calculated, which is a reproducibility gap rather than circularity. Self-citations to the authors' prior work, such as R-TOSS [23], appear as baseline comparison and background, not as load-bearing justification for the central claims. Bugs in the pseudocode, such as Algorithm 4 using temp_array before assignment, are correctness issues and do not constitute circular reasoning.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central method rests on two unverified pillars: an unstated on-device latency/energy model and an unspecified KITTI evaluation protocol. The ES weights and bit-search range are hand-set free parameters. No invented entities are introduced.

free parameters (4)
  • Efficiency score weights alpha, beta, gamma = alpha=0.3, beta=0.4, gamma=0.3
    Hand-set before experiments to favor latency; no sensitivity analysis is reported, and the weights change which kernels are selected (Section IV.C, Eq. 2).
  • Quantization bit search range = 4 to 16 bits
    Search space for mixed-precision quantizer; the per-kernel bit assignments are not reported, so the actual compressed precision is unknown.
  • Nonzero weights per 3x3 pattern (n) = HCK: 2, LCK: 3
    Controls sparsity aggressiveness between the two UPAQ variants; the paper does not explore other values.
  • 1x1 kernel grouping size (k) = Undefined in the paper
    Algorithm 5 reshapes flattened 1x1 weights into k x k blocks, but k is never defined or set, making the 1x1 compression procedure underspecified.
assumptions (5)
  • domain assumption Latency and energy of the whole compressed model can be estimated at per-kernel granularity during optimization.
    Algorithms 4 and 5 invoke calculate_ES after each kernel update, but Section IV.C gives no latency/energy model, no equations, and no calibration, so the central optimizer rests on an unverified modeling step.
  • ad hoc to paper The four handpicked pattern families (main diagonal, anti-diagonal, row, column) are sufficient to preserve accuracy.
    Algorithm 2 randomly selects from a fixed list with no coverage analysis or evidence that these patterns span the useful mask space.
  • domain assumption SQNR is an adequate proxy for detection mAP when comparing candidate kernels.
    Eq. (2) uses SQNR as the accuracy term of the efficiency score, but the paper provides no correlation analysis between SQNR and final mAP.
  • domain assumption The 80:10:10 KITTI split and the mAP metric follow a standard, internally consistent evaluation protocol.
    Section V.A states the split but omits the KITTI difficulty level, IoU threshold, 2D/BEV/3D mAP kind, and interpolation method; without this the base-vs-compressed accuracy comparison is unverifiable.
  • domain assumption The re-implemented baseline frameworks faithfully reproduce the methods and performance of their original papers.
    The authors compare against Ps&Qs, CLIP-Q, R-TOSS, and LIDAR-PTQ without releasing code or documenting their implementations, so the fairness of the SOTA comparison cannot be audited.

how reviews work

0 comments
Cite this review

Pith. "Pith review of UPAQ: A Framework for Real-Time and Energy-Efficient 3D Object Detection in Autonomous Vehicles." pith.science (2026). https://pith.science/paper/HA6LIMR3

@misc{pith2026250104213,
  author       = {Pith},
  title        = {Pith review of: UPAQ: A Framework for Real-Time and Energy-Efficient 3D Object Detection in Autonomous Vehicles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HA6LIMR3}},
  note         = {Machine review of arXiv:2501.04213}
}
read the original abstract

To enhance perception in autonomous vehicles (AVs), recent efforts are concentrating on 3D object detectors, which deliver more comprehensive predictions than traditional 2D object detectors, at the cost of increased memory footprint and computational resource usage. We present a novel framework called UPAQ, which leverages semi-structured pattern pruning and quantization to improve the efficiency of LiDAR point-cloud and camera-based 3D object detectors on resource-constrained embedded AV platforms. Experimental results on the Jetson Orin Nano embedded platform indicate that UPAQ achieves up to 5.62x and 5.13x model compression rates, up to 1.97x and 1.86x boost in inference speed, and up to 2.07x and 1.87x reduction in energy consumption compared to state-of-the-art model compression frameworks, on the Pointpillar and SMOKE models respectively.

Figures

Figures reproduced from arXiv: 2501.04213 by the authors.

Figure 3
Figure 3. An overview of the proposed UPAQ optimization framework. A. Pre-processing Stage In this first stage, we begin by calculating the computational graph of the pretrained model M and utilize the depth-first search (DFS) algorithm to determine the computation paths or connected layers. A root layer may be shared among multiple leaf layers, which we leverage to significantly lower the computational cost associated with o… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 36 canonical work pages

  1. [1]

    Automated Driving Systems, NHTSA, https://www.nhtsa.gov/ [last accessed on: 10/15/2024]

    work page 2024

  2. [2]

    Object detection in autonomous vehicles: Status and open challenges

    A. Balasubramaniam, S. Pasricha , "Object detection in autonomous vehicles: Status and open challenges." arXiv preprint arXiv:2201.07706 ,2022

    arXiv 2022

  3. [3]

    Advanced Driver Assistance Systems: A Path Toward Autonomous Vehicles

    V. Kukkala, J. Tunnell, S. Pasricha , "Advanced Driver Assistance Systems: A Path Toward Autonomous Vehicles", IEEE Consumer Electronics, vol. 7, no. 5, Sept 2018

    work page 2018

  4. [4]

    Machine Learning and Optimization Techniques for Automotive Cyber -Physical Systems

    V. Kukkala, S. Pasricha, "Machine Learning and Optimization Techniques for Automotive Cyber -Physical Systems ", Springer Nature Publishers, 2023

    work page 2023

  5. [5]

    Object detection based on lightweight YOLOX for autonomous driving

    H. Qiyi, et al. "Object detection based on lightweight YOLOX for autonomous driving." Sensors 23.17 ,2023

    work page 2023

  6. [6]

    Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks

    H. Torsten, et al . "Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks." The Journal of Machine Learning Research 22.1 ,2021

    work page 2021

  7. [7]

    Pruning vs Quantization: Which is Better?

    K. Andrey, et al . "Pruning vs Quantization: Which is Better?." arXiv preprint arXiv:2307.02973 ,2023

    arXiv 2023

  8. [8]

    Fast R-CNN,

    R. Girshick, “Fast R-CNN,” IEEE ICCV, 2015

    work page 2015

Show all 41 references
  1. [9]

    Faster R-CNN: an Approach to Real -Time Object Detection,

    R. Gavrilescu, et al., “Faster R-CNN: an Approach to Real -Time Object Detection,” IEEE EPE, 2018

    2018

  2. [10]

    Focal Loss for Dense Object Detection,

    T. Y. Lin, et al. , “Focal Loss for Dense Object Detection,” arXiv:1708.02002, 2018

    2018 arXiv

  3. [11]

    On the performance of one -stage and two -stage object detectors in autonomous vehicles using camera data

    C.G, Manuel, et al. "On the performance of one -stage and two -stage object detectors in autonomous vehicles using camera data." Remote Sensing 13.1 (2020): 89

    2020

  4. [12]

    YOLOX: Exceeding YOLO Series in 2021,

    Z. Ge, et al. , “YOLOX: Exceeding YOLO Series in 2021,” arXiv:2107.08430, 2021

    2021 arXiv

  5. [13]

    Pointnet: Deep learning on point sets for 3d classification and segmentation

    Q. Charles, et al. "Pointnet: Deep learning on point sets for 3d classification and segmentation." Proceedings of the IEEE conference on computer vision and pattern recognition, 2017

    2017

  6. [14]

    Pointpillars: Fast encoders for object detection from point clouds

    H. Alex, et al. "Pointpillars: Fast encoders for object detection from point clouds." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019

    2019

  7. [15]

    Objects are different: Flexible monocular 3d object detection

    Z. Yunpeng, et al. "Objects are different: Flexible monocular 3d object detection." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021

    2021

  8. [16]

    Smoke: Single-stage monocular 3d object detection via keypoint estimation

    L. Zechen, et al. "Smoke: Single-stage monocular 3d object detection via keypoint estimation." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , 2020

    2020

  9. [17]

    Second: Sparsely embedded convolutional detection

    Y. Yan, et al. "Second: Sparsely embedded convolutional detection." Sensors 18.10, 2018

    2018

  10. [18]

    Focal sparse convolutional networks for 3d object detection

    C. Yukang, et al. "Focal sparse convolutional networks for 3d object detection." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 2022

    2022

  11. [19]

    Virtual sparse convolution for multimodal 3d object detection

    W. Hai, et al. "Virtual sparse convolution for multimodal 3d object detection." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 2023

    2023

  12. [20]

    Ps and qs : Quantization-aware pruning for efficient low latency neural network inference

    H. Benjamin, et al. "Ps and qs : Quantization-aware pruning for efficient low latency neural network inference." Frontiers in Artificial Intelligence 4, 2021

    2021

  13. [21]

    Clip-q: Deep network compression learning by in - parallel pruning -quantization

    T. Frederick, et al. "Clip-q: Deep network compression learning by in - parallel pruning -quantization." Proceedings of the IEEE conference on computer vision and pattern recognition , 2018

    2018

  14. [22]

    Lidar-ptq: Post-training quantization for point cloud 3d object detection

    Z., Sifan, et al. "Lidar-ptq: Post-training quantization for point cloud 3d object detection." arXiv preprint arXiv:2401.15865 (2024)

    2024 arXiv

  15. [23]

    R-TOSS: A framework for real-time object detection using semi -structured pruning

    A. Balasubramaniam, F. Sunny, S. Pasricha , "R-TOSS: A framework for real-time object detection using semi -structured pruning." 2023 60th ACM/IEEE Design Automation Conference (DAC). IEEE , 2023

    2023

  16. [24]

    Inducing and Exploiting Activation Sparsity for Fast Inference on Deep Neural Networks,

    M. Kurtz et al. , “Inducing and Exploiting Activation Sparsity for Fast Inference on Deep Neural Networks,” in ICML, 2020

    2020

  17. [25]

    Importance estimation for neural network pruning,

    P. Molchanov, et al., "Importance estimation for neural network pruning,” in IEEE CVPR, 2019

    2019

  18. [26]

    Pruning neural networks without any data by iteratively conserving synaptic flow,

    H. Tanaka, et al., "Pruning neural networks without any data by iteratively conserving synaptic flow," in NeurIPS ,2020

    2020

  19. [27]

    Edge devices object detection by filter pruning,

    V. Crescitelli, et. al., “Edge devices object detection by filter pruning,” IEEE ETFA, 2021

    2021

  20. [28]

    Localization -aware channel pruning for object detection,

    X. Zihao, et al. , “Localization -aware channel pruning for object detection,” Neurocomputing vol. 403, 2020

    2020

  21. [29]

    TensorRT-Based Framework and Optimization Methodology for Deep Learning Inference on Jetson Boards

    J, EunJin, et al. "TensorRT-Based Framework and Optimization Methodology for Deep Learning Inference on Jetson Boards." ACM Transactions on Embedded Computing Systems , 2022

    2022

  22. [30]

    Patdnn: Achieving real -time dnn execution on mobile devices with pattern -based weight pruning,

    N. Wei, et al. , "Patdnn: Achieving real -time dnn execution on mobile devices with pattern -based weight pruning," in International Conference on Architectural Support for Programming Languages and Operating Systems, 2020

    2020

  23. [31]

    Loss aware post -training quantization

    N. Yury, et al. "Loss aware post -training quantization." Machine Learning 110.11-12 ,2021

    2021

  24. [32]

    A survey of quantization methods for efficient neural network inference

    G. Amir, et al. "A survey of quantization methods for efficient neural network inference." Low-Power Computer Vision. Chapman and Hall/CRC, 2022. 291-326

    2022

  25. [33]

    KITTI dataset; Available on: https://www.cvlibs.net/datasets/kitti/ [last accessed on: 10/15/2024]

    2024

  26. [34]

    https://docs.python.org/3/library/copy.html [last accessed on: 11/13/2024]

    Shallow and deep copy operations - Python documentation,” docs.python.org. https://docs.python.org/3/library/copy.html [last accessed on: 11/13/2024]

    2024

  27. [35]

    Quantization and training of neural networks for efficient integer-arithmetic-only inference

    J. Benoit, et al. "Quantization and training of neural networks for efficient integer-arithmetic-only inference ." Proceedings of the IEEE conference on computer vision and pattern recognition , 2018

    2018

  28. [36]

    NVpower at master · wildkid1024/NVpower,

    wildkid1024, “NVpower at master · wildkid1024/NVpower,” GitHub,

  29. [37]

    Roadmap for Cybersecurity in Autonomous Vehicles

    V. K. Kukkala, S. V. Thiruloga, and S. Pasricha, "Roadmap for Cybersecurity in Autonomous Vehicles ", Vol. 11, Iss. 6, pp. 13-23, IEEE Consumer Electronics , Nov 2022

    2022

  30. [38]

    LATTE: LSTM Self - Attention based Anomaly Detection in Embedded Automotive Platforms

    V. K. Kukkala, S. V. Thiruloga, and S. Pasricha, "LATTE: LSTM Self - Attention based Anomaly Detection in Embedded Automotive Platforms", ACM Transactions on Embedded Computing Systems (TECS), vol 20, Issue 5s, Oct 2021

    2021

  31. [39]

    VESPA: Optimizing Heterogeneous Sensor Placement and Orientation for Autonomous Vehicles

    J. Dey, W. Taylor, S. Pasricha, "VESPA: Optimizing Heterogeneous Sensor Placement and Orientation for Autonomous Vehicles ", IEEE Consumer Electronics, 10(2), Mar 2021

    2021

  32. [40]

    Co-Optimizing Sensing and Deep Machine Learning in Automotive Cyber-Physical Systems

    J. Dey, S. Pasricha, "Co-Optimizing Sensing and Deep Machine Learning in Automotive Cyber-Physical Systems", IEEE Euromicro Conference on Digital Systems Design, 2022

    2022

  33. [2020]

    28, 2024]

    https://github.com/wildkid1024/NVpower/tree/master/NVpower [last accessed Dec. 28, 2024]

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.