pith. sign in

arxiv: 2511.01553 · v2 · submitted 2025-11-03 · 💻 cs.LG · cs.AI· cs.DC· cs.NE

Online Continual Learning on Intel Loihi 2 via a Co-designed Spiking Neural Network

Elvin Hajizada , Danielle Rager , Timothy Shea , Leobardo Campos-Macias , Andreas Wild , Eyke H\"ullermeier , Yulia Sandamirskaya , Mike Davies This is my paper

Pith reviewed 2026-05-18 01:06 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.DCcs.NE
keywords continual learningspiking neural networkneuromorphic computingLoihi 2online learningedge AIrehearsal-freefew-shot learning
0
0 comments X

The pith

CLP-SNN on Loihi 2 matches replay accuracy without rehearsal while using 113x lower latency and 6600x lower energy than edge GPUs.

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

The paper introduces CLP-SNN, a spiking neural network built for online continual learning on neuromorphic chips. It combines a self-normalizing local learning rule with a spike-driven neural state machine so the network can adapt to new classes and shifting data streams on the device itself. Because the approach needs no stored examples or replay buffer, it avoids the memory and power costs of traditional rehearsal methods. When run on Intel Loihi 2, the system reaches the same few-shot accuracy as replay baselines on OpenLORIS yet finishes each update in 0.33 ms and 0.05 mJ, far below GPU numbers. The reported gains split between the algorithm's sparse event-driven design and the hardware's native support for graded spikes and on-chip plasticity.

Core claim

CLP-SNN is a spiking neural network that incorporates a self-normalizing local learning rule and a spike-driven neural state machine for autonomous on-chip learning. Implemented on Intel's Loihi 2 neuromorphic processor, it matches the accuracy of replay-based methods on OpenLORIS few-shot experiments without any rehearsal. It achieves 113x lower latency and 6,600x lower energy than the strongest edge-GPU baseline. The efficiency improvements arise from both algorithmic efficiency on the same GPU and neuromorphic hardware co-design exploiting event-driven learning and sparse graded-spike communication.

What carries the argument

The self-normalizing local learning rule combined with the spike-driven neural state machine, which together enable rehearsal-free adaptation and sparse, event-driven computation directly on neuromorphic hardware.

If this is right

  • CLP-SNN enables on-device adaptation to non-stationary streams without storing or replaying past examples.
  • Neuromorphic co-design can break traditional accuracy-efficiency trade-offs for edge continual learning.
  • Event-driven learning and graded-spike communication account for the majority of the measured energy reduction.
  • The same architecture supports autonomous on-chip learning under strict power constraints typical of edge devices.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The memory savings from eliminating rehearsal could let larger networks fit on the same neuromorphic chip.
  • Similar algorithm-hardware pairings might be tested on other event-driven processors for robotics or sensor networks.
  • If the local rule generalizes, it could reduce dependence on cloud retraining for always-on edge systems.

Load-bearing premise

The self-normalizing local learning rule combined with the spike-driven neural state machine prevents catastrophic forgetting in non-stationary data streams without any rehearsal or stored examples.

What would settle it

A test on a longer or more abrupt sequence of OpenLORIS classes where CLP-SNN accuracy falls more than a few percent below a replay baseline while using the same number of training examples.

Figures

Figures reproduced from arXiv: 2511.01553 by Andreas Wild, Danielle Rager, Elvin Hajizada, Eyke H\"ullermeier, Leobardo Campos-Macias, Mike Davies, Timothy Shea, Yulia Sandamirskaya.

Figure 1
Figure 1. Figure 1: From dense global updates to event-driven local learning on neuromorphic chip Loihi 2, as an efficient solution to online continual learning. a Catastrophic forgetting happens when learning tasks (e.g. objects) are presented sequentially in non-i.i.d data streams. Online Continual Learning (OCL) is such a setting, where inference and learning occur per sample. b (Left) Global learning with backpropagation … view at source ↗
Figure 2
Figure 2. Figure 2: The proposed spiking neural network (SNN) architecture for the CLP algorithm [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Few-shot online continual learning experiments with OpenLORIS dataset. a [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
read the original abstract

AI systems on edge devices require online continual learning -- adapting to non-stationary streams and unfamiliar classes without catastrophic forgetting -- under strict power constraints. We present CLP-SNN, a spiking neural network with a self-normalizing local learning rule and a spike-driven neural state machine for autonomous on-chip learning, implemented on Intel's Loihi 2 neuromorphic processor. On OpenLORIS few-shot experiments, CLP-SNN matches replay-based accuracy rehearsal-free. On Loihi 2, CLP-SNN achieves 113x lower latency (0.33 ms vs. 37.3 ms) and 6,600x lower energy (0.05 mJ vs. 333 mJ) than the strongest edge-GPU baseline. This gain decomposes into algorithmic efficiency (~14.5x latency, ~22.6x energy on the same GPU) and neuromorphic hardware co-design (~7.8x latency, ~295x energy) exploiting event-driven learning and sparse graded-spike communication. We show that co-designed brain-inspired algorithms and neuromorphic hardware can break traditional accuracy-efficiency trade-offs in edge AI.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces CLP-SNN, a spiking neural network architecture that incorporates a self-normalizing local learning rule and a spike-driven neural state machine to enable rehearsal-free online continual learning on the Intel Loihi 2 neuromorphic processor. On OpenLORIS few-shot experiments, the approach is reported to match the accuracy of replay-based methods while remaining strictly rehearsal-free. On Loihi 2 hardware, it achieves 113x lower latency (0.33 ms vs. 37.3 ms) and 6,600x lower energy (0.05 mJ vs. 333 mJ) relative to the strongest edge-GPU baseline, with the gains decomposed into algorithmic contributions (~14.5x latency, ~22.6x energy) and neuromorphic hardware co-design (~7.8x latency, ~295x energy) arising from event-driven learning and sparse graded-spike communication.

Significance. If the experimental results hold under scrutiny, the work would be significant for edge AI applications by demonstrating that brain-inspired co-design can simultaneously achieve accuracy parity with rehearsal-based continual learning and orders-of-magnitude efficiency improvements on neuromorphic hardware. The explicit decomposition of latency and energy gains into algorithmic versus hardware factors provides a useful template for future neuromorphic algorithm development.

major comments (2)
  1. [Abstract and Results] Abstract and Results: The central claim of accuracy parity with replay-based methods on OpenLORIS few-shot experiments is presented without error bars, dataset split details, or ablation studies on the self-normalizing rule and state machine; this makes it difficult to evaluate the reliability of the rehearsal-free performance assertion and whether the local learning rule truly prevents catastrophic forgetting in non-stationary streams.
  2. [Methods] Methods: The description of the spike-driven neural state machine and self-normalizing local learning rule is high-level in the provided text; without explicit equations or pseudocode showing how normalization and state transitions maintain stability across class-incremental streams, it is hard to assess the internal mechanism supporting the no-rehearsal claim.
minor comments (1)
  1. [Abstract] Abstract: The strongest edge-GPU baseline is not named when reporting the 113x latency and 6,600x energy figures, which would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments on our manuscript. We address each major point below and have revised the manuscript accordingly to improve clarity and provide additional supporting details.

read point-by-point responses

  1. Referee: [Abstract and Results] Abstract and Results: The central claim of accuracy parity with replay-based methods on OpenLORIS few-shot experiments is presented without error bars, dataset split details, or ablation studies on the self-normalizing rule and state machine; this makes it difficult to evaluate the reliability of the rehearsal-free performance assertion and whether the local learning rule truly prevents catastrophic forgetting in non-stationary streams.

    Authors: We agree that additional statistical details and ablations would strengthen the presentation. The reported accuracies were obtained by averaging over five independent runs with different random seeds; we have now added error bars (standard deviation) to the relevant figures and tables in the revised manuscript. We have also expanded Section 4.1 to explicitly describe the OpenLORIS few-shot dataset splits and class-incremental protocol. New ablation studies have been added in Section 5.3 that remove the self-normalizing term or the neural state machine individually; these confirm that both components are required to achieve accuracy parity with replay baselines while avoiding catastrophic forgetting, with the full CLP-SNN model showing stable performance across non-stationary streams. revision: yes

  2. Referee: [Methods] Methods: The description of the spike-driven neural state machine and self-normalizing local learning rule is high-level in the provided text; without explicit equations or pseudocode showing how normalization and state transitions maintain stability across class-incremental streams, it is hard to assess the internal mechanism supporting the no-rehearsal claim.

    Authors: We acknowledge that the original methods description was concise. In the revised manuscript we have expanded Sections 3.2 and 3.3 with the full mathematical formulation of the self-normalizing local learning rule, including the homeostatic normalization term that bounds weight updates and prevents divergence. We have also inserted Algorithm 1, which provides pseudocode for the spike-driven neural state machine, detailing how graded spikes trigger state transitions and enable autonomous class-incremental adaptation without external rehearsal. These additions make the stability mechanism explicit and directly support the rehearsal-free claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity; results are direct hardware measurements

full rationale

The paper presents CLP-SNN as a hardware implementation on Loihi 2, with performance claims (latency, energy, accuracy on OpenLORIS) framed as experimental measurements rather than derived predictions or first-principles results. The self-normalizing local learning rule and spike-driven state machine are design elements validated through rehearsal-free few-shot experiments, not quantities that reduce to fitted inputs or self-referential equations by construction. No load-bearing self-citations, uniqueness theorems, or ansatzes are invoked in a way that collapses the central claims. The derivation chain consists of algorithmic co-design and hardware benchmarking, which are externally falsifiable via replication on the described platform. This is a standard non-circular experimental report.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the effectiveness of the newly introduced self-normalizing local learning rule and spike-driven state machine for preventing forgetting without rehearsal; these are presented as domain innovations rather than derived from first principles.

axioms (1)
  • domain assumption Local self-normalizing learning rules suffice to avoid catastrophic forgetting in non-stationary streams without rehearsal.
    Invoked to justify rehearsal-free operation on OpenLORIS few-shot tasks.
invented entities (1)
  • CLP-SNN architecture no independent evidence
    purpose: Spiking network with integrated local learning rule and spike-driven state machine for on-chip continual learning
    The full architecture and its components are introduced in this work.

pith-pipeline@v0.9.0 · 5770 in / 1460 out tokens · 51370 ms · 2026-05-18T01:06:49.374885+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages

  1. [1]

    Metaplasticity: tuning synapses and networks for plasticity.Nature Re- views Neuroscience, 9(5):387–387, 2008

    Wickliffe C Abraham. Metaplasticity: tuning synapses and networks for plasticity.Nature Re- views Neuroscience, 9(5):387–387, 2008

    work page 2008

  2. [2]

    Chameleon: Dual memory replay for online continual learning on edge devices.IEEE Trans- actions on Computer-Aided Design of Integrated Circuits and Systems, 43(6):1663–1676, 2023

    Shivam Aggarwal, Kuluhan Binici, and Tulika Mi- tra. Chameleon: Dual memory replay for online continual learning on edge devices.IEEE Trans- actions on Computer-Aided Design of Integrated Circuits and Systems, 43(6):1663–1676, 2023

    work page 2023

  3. [3]

    Computational influence of adult neurogenesis on memory encoding.Neuron, 61(2):187–202, 2009

    James B Aimone, Janet Wiles, and Fred H Gage. Computational influence of adult neurogenesis on memory encoding.Neuron, 61(2):187–202, 2009

    work page 2009

  4. [4]

    Computational principles of synaptic memory consolidation.Na- ture neuroscience, 19(12):1697–1706, 2016

    Marcus K Benna and Stefano Fusi. Computational principles of synaptic memory consolidation.Na- ture neuroscience, 19(12):1697–1706, 2016

    work page 2016

  5. [5]

    Online con- tinual learning: A systematic literature review of approaches, challenges, and benchmarks.arXiv preprint arXiv:2501.04897, 2025

    Seyed Amir Bidaki, Amir Mohammadkhah, Kiyan Rezaee, Faeze Hassani, Sadegh Eskandari, Maziar Salahi, and Mohammad M Ghassemi. Online con- tinual learning: A systematic literature review of approaches, challenges, and benchmarks.arXiv preprint arXiv:2501.04897, 2025

    work page arXiv 2025

  6. [6]

    Continual learning about objects in the wild: An interactive approach

    Dan Bohus, Sean Andrist, Ashley Feniello, Nick Saw, and Eric Horvitz. Continual learning about objects in the wild: An interactive approach. In Proceedings of the 2022 International Conference on Multimodal Interaction, pages 476–486, 2022

    work page 2022

  7. [7]

    Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901, 2020

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sas- try, Amanda Askell, et al. Language models are few-shot learners.Advances in neural information processing systems, 33:1877–1901, 2020

    work page 1901

  8. [8]

    Loihi: A neuromorphic manycore processor with on-chip learning.IEEE Micro, 38:82–99, 2018

    Mike Davies, Narayan Srinivasa, Tsung Han Lin, Gautham Chinya, Yongqiang Cao, Sri Harsha Cho- day, Georgios Dimou, Prasad Joshi, Nabil Imam, Shweta Jain, Yuyun Liao, Chit Kwan Lin, An- drew Lines, Ruokun Liu, Deepak Mathaikutty, Steven McCoy, Arnab Paul, Jonathan Tse, Gu- ruguhanathan Venkataramanan, Yi Hsin Weng, Andreas Wild, Yoonseok Yang, and Hong Wan...

    work page 2018

  9. [9]

    Advancing neuromorphic computing with loihi: A survey of results and outlook.Proceedings of the IEEE, 109(5):911–934, 2021

    Mike Davies, Andreas Wild, Garrick Orchard, Yu- lia Sandamirskaya, Gabriel A Fonseca Guerra, Prasad Joshi, Philipp Plank, and Sumedh R Ris- bud. Advancing neuromorphic computing with loihi: A survey of results and outlook.Proceedings of the IEEE, 109(5):911–934, 2021

    work page 2021

  10. [10]

    The evolving view of replay and its functions in wake and sleep.Sleep Advances, 1(1):zpab002, 2020

    Graham Findlay, Giulio Tononi, and Chiara Cirelli. The evolving view of replay and its functions in wake and sleep.Sleep Advances, 1(1):zpab002, 2020

    work page 2020

  11. [11]

    Cambridge University Press, 2014

    Wulfram Gerstner, Werner M Kistler, Richard Naud, and Liam Paninski.Neuronal dynamics: From single neurons to networks and models of cognition. Cambridge University Press, 2014

    work page 2014

  12. [12]

    Interactive continual learning for robots: a neuro- morphic approach

    Elvin Hajizada, Patrick Berggold, Massimiliano Iacono, Arren Glover, and Yulia Sandamirskaya. Interactive continual learning for robots: a neuro- morphic approach. InProceedings of the Interna- tional Conference on Neuromorphic Systems 2022, pages 1–10, 2022

    work page 2022

  13. [13]

    Continual learning for au- tonomous robots: A prototype-based approach

    Elvin Hajizada, Balachandran Swaminathan, and Yulia Sandamirskaya. Continual learning for au- tonomous robots: A prototype-based approach. In 2024 IEEE/RSJ International Conference on Intel- ligent Robots and Systems (IROS), pages 13988– 13995, 2024

    work page 2024

  14. [14]

    Life- long machine learning with deep streaming lin- ear discriminant analysis

    Tyler L Hayes and Christopher Kanan. Life- long machine learning with deep streaming lin- ear discriminant analysis. InProceedings of the IEEE/CVF conference on computer vision and pat- tern recognition workshops, pages 220–221, 2020

    work page 2020

  15. [15]

    Online continual learning for embedded devices.arXiv preprint arXiv:2203.10681, 2022

    Tyler L Hayes and Christopher Kanan. Online continual learning for embedded devices.arXiv preprint arXiv:2203.10681, 2022

    work page arXiv 2022

  16. [16]

    Psychology press, 2005

    Donald Olding Hebb.The organization of behavior: A neuropsychological theory. Psychology press, 2005

    work page 2005

  17. [17]

    Neurogenesis dynamics-inspired spiking neural network train- ing acceleration

    Shaoyi Huang, Haowen Fang, Kaleel Mahmood, Bowen Lei, Nuo Xu, Bin Lei, Yue Sun, Dongkuan Xu, Wujie Wen, and Caiwen Ding. Neurogenesis dynamics-inspired spiking neural network train- ing acceleration. In2023 60th ACM/IEEE Design Automation Conference (DAC), pages 1–6. IEEE, 2023

    work page 2023

  18. [18]

    The role of experience in prioritiz- ing hippocampal replay.Nature Communications, 14(1):8157, 2023

    Marta Huelin Gorriz, Masahiro Takigawa, and Daniel Bendor. The role of experience in prioritiz- ing hippocampal replay.Nature Communications, 14(1):8157, 2023

    work page 2023

  19. [19]

    Rapid online learning and robust recall in a neuromorphic olfac- tory circuit.Nature Machine Intelligence, 2(3):181– 191, 2020

    Nabil Imam and Thomas A Cleland. Rapid online learning and robust recall in a neuromorphic olfac- tory circuit.Nature Machine Intelligence, 2(3):181– 191, 2020

    work page 2020

  20. [20]

    Lava: A software framework for neuromorphic computing, 2021

    Intel Corporation. Lava: A software framework for neuromorphic computing, 2021. Accessed: [Date]

    work page 2021

  21. [21]

    How in- hibition shapes cortical activity.Neuron, 72(2):231– 243, 2011

    Jeffry S Isaacson and Massimo Scanziani. How in- hibition shapes cortical activity.Neuron, 72(2):231– 243, 2011

    work page 2011

  22. [22]

    How in- hibition shapes cortical activity.Neuron, 72(2):231– 243, 2011

    Jeffry S Isaacson and Massimo Scanziani. How in- hibition shapes cortical activity.Neuron, 72(2):231– 243, 2011. 14

    work page 2011

  23. [23]

    A review of open-world learning and steps toward open-world learning without la- bels.arXiv preprint arXiv:2011.12906, 2020

    Mohsen Jafarzadeh, Akshay Raj Dhamija, Steve Cruz, Chunchun Li, Touqeer Ahmad, and Ter- rance E Boult. A review of open-world learning and steps toward open-world learning without la- bels.arXiv preprint arXiv:2011.12906, 2020

    work page arXiv 2011

  24. [24]

    Towards open world object detection

    KJ Joseph, Salman Khan, Fahad Shahbaz Khan, and Vineeth N Balasubramanian. Towards open world object detection. InProceedings of the IEEE/CVF conference on computer vision and pat- tern recognition, pages 5830–5840, 2021

    work page 2021

  25. [25]

    Training machine learning models at the edge: A survey

    Aymen Rayane Khouas, Mohamed Reda Bouad- jenek, Hakim Hacid, and Sunil Aryal. Training machine learning models at the edge: A survey. arXiv preprint arXiv:2403.02619, 2024

    work page arXiv 2024

  26. [26]

    On the stability- plasticity dilemma of class-incremental learning

    Dongwan Kim and Bohyung Han. On the stability- plasticity dilemma of class-incremental learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20196–20204, 2023

    work page 2023

  27. [27]

    Biological underpinnings for lifelong learning ma- chines.Nature Machine Intelligence, 4(3):196–210, 2022

    Dhireesha Kudithipudi, Mario Aguilar-Simon, Jonathan Babb, Maxim Bazhenov, Douglas Black- iston, Josh Bongard, Andrew P Brna, Suraj Chakravarthi Raja, Nick Cheney, Jeff Clune, et al. Biological underpinnings for lifelong learning ma- chines.Nature Machine Intelligence, 4(3):196–210, 2022

    work page 2022

  28. [28]

    Training on the edge: The why and the how

    Navjot Kukreja, Alena Shilova, Olivier Beaumont, Jan Huckelheim, Nicola Ferrier, Paul Hovland, and Gerard Gorman. Training on the edge: The why and the how. In2019 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), pages 899–903. IEEE, 2019

    work page 2019

  29. [29]

    Rilod: Near real-time incremental learning for object detection at the edge

    Dawei Li, Serafettin Tasci, Shalini Ghosh, Jingwen Zhu, Junting Zhang, and Larry Heck. Rilod: Near real-time incremental learning for object detection at the edge. InProceedings of the 4th ACM/IEEE Symposium on Edge Computing, pages 113–126, 2019

    work page 2019

  30. [30]

    Backpropa- gation and the brain.Nature Reviews Neuroscience, 21(6):335–346, 2020

    Timothy P Lillicrap, Adam Santoro, Luke Marris, Colin J Akerman, and Geoffrey Hinton. Backpropa- gation and the brain.Nature Reviews Neuroscience, 21(6):335–346, 2020

    work page 2020

  31. [31]

    On-device training under 256kb memory.Advances in Neural Information Processing Systems, 35:22941–22954, 2022

    Ji Lin, Ligeng Zhu, Wei-Ming Chen, Wei-Chen Wang, Chuang Gan, and Song Han. On-device training under 256kb memory.Advances in Neural Information Processing Systems, 35:22941–22954, 2022

    work page 2022

  32. [32]

    Catas- trophic interference in connectionist networks: The sequential learning problem

    Michael McCloskey and Neal J Cohen. Catas- trophic interference in connectionist networks: The sequential learning problem. InPsychology of learning and motivation, volume 24, pages 109–

    work page

  33. [33]

    Continual learning with neuromor- phic computing: Theories, methods, and applica- tions.arXiv preprint arXiv:2410.09218, 2024

    Mishal Fatima Minhas, Rachmad Vidya Wicaksana Putra, Falah Awwad, Osman Hasan, and Muham- mad Shafique. Continual learning with neuromor- phic computing: Theories, methods, and applica- tions.arXiv preprint arXiv:2410.09218, 2024

    work page arXiv 2024

  34. [34]

    A survey and perspective on neuromorphic continual learning systems.Frontiers in Neuroscience, 17:1149410, 2023

    Richa Mishra and Manan Suri. A survey and perspective on neuromorphic continual learning systems.Frontiers in Neuroscience, 17:1149410, 2023

    work page 2023

  35. [35]

    Simplified neuron model as a princi- pal component analyzer.Journal of mathematical biology, 15:267–273, 1982

    Erkki Oja. Simplified neuron model as a princi- pal component analyzer.Journal of mathematical biology, 15:267–273, 1982

    work page 1982

  36. [36]

    Efficient neuromorphic signal processing with loihi

    Garrick Orchard, E Paxon Frady, Daniel Ben Dayan Rubin, Sophia Sanborn, Sumit Bam Shrestha, Friedrich T Sommer, and Mike Davies. Efficient neuromorphic signal processing with loihi

    work page

  37. [37]

    IEEE, 2021

    In2021 IEEE Workshop on Signal Processing Systems (SiPS), pages 254–259. IEEE, 2021

    work page 2021

  38. [38]

    Challenges in deploying machine learn- ing: a survey of case studies.ACM computing surveys, 55(6):1–29, 2022

    Andrei Paleyes, Raoul-Gabriel Urma, and Neil D Lawrence. Challenges in deploying machine learn- ing: a survey of case studies.ACM computing surveys, 55(6):1–29, 2022

    work page 2022

  39. [39]

    Continual learn- ing at the edge: Real-time training on smartphone devices.arXiv preprint arXiv:2105.13127, 2021

    Lorenzo Pellegrini, Vincenzo Lomonaco, Gabriele Graffieti, and Davide Maltoni. Continual learn- ing at the edge: Real-time training on smartphone devices.arXiv preprint arXiv:2105.13127, 2021

    work page arXiv 2021

  40. [40]

    Mlsysbook.ai: Principles and practices of machine learning systems engi- neering

    Vijay Janapa Reddi. Mlsysbook.ai: Principles and practices of machine learning systems engi- neering. In2024 International Conference on Hardware/Software Codesign and System Synthesis (CODES+ ISSS), pages 41–42. IEEE, 2024

    work page 2024

  41. [41]

    Tinyol: Tinyml with online-learning on microcon- trollers

    Haoyu Ren, Darko Anicic, and Thomas A Runkler. Tinyol: Tinyml with online-learning on microcon- trollers. In2021 international joint conference on neural networks (IJCNN), pages 1–8. IEEE, 2021

    work page 2021

  42. [42]

    Wandering within a world: Online contextualized few-shot learning

    Mengye Ren, Michael L Iuzzolino, Michael C Mozer, and Richard S Zemel. Wandering within a world: Online contextualized few-shot learning. arXiv preprint arXiv:2007.04546, 2020

    work page arXiv 2007

  43. [43]

    Opportunities for neuromorphic computing algorithms and applications.Nature Computational Science, 2(1):10–19, 2022

    Catherine D Schuman, Shruti R Kulkarni, Maryam Parsa, J Parker Mitchell, Prasanna Date, and Bill Kay. Opportunities for neuromorphic computing algorithms and applications.Nature Computational Science, 2(1):10–19, 2022

    work page 2022

  44. [44]

    Continual learning for real-world autonomous systems: Al- gorithms, challenges and frameworks.Journal of Intelligent & Robotic Systems, 105(1):9, 2022

    Khadija Shaheen, Muhammad Abdullah Hanif, Os- man Hasan, and Muhammad Shafique. Continual learning for real-world autonomous systems: Al- gorithms, challenges and frameworks.Journal of Intelligent & Robotic Systems, 105(1):9, 2022

    work page 2022

  45. [45]

    Openloris-object: A robotic vision dataset and benchmark for lifelong deep learning

    Qi She, Fan Feng, Xinyue Hao, Qihan Yang, Chuanlin Lan, Vincenzo Lomonaco, Xuesong Shi, 15 Zhengwei Wang, Yao Guo, Yimin Zhang, et al. Openloris-object: A robotic vision dataset and benchmark for lifelong deep learning. In2020 IEEE international conference on robotics and au- tomation (ICRA), pages 4767–4773. IEEE, 2020

    work page 2020

  46. [46]

    Efficient video and audio processing with loihi 2

    Sumit Bam Shrestha, Jonathan Timcheck, Paxon Frady, Leobardo Campos-Macias, and Mike Davies. Efficient video and audio processing with loihi 2. InICASSP 2024-2024 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 13481–13485. IEEE, 2024

    work page 2024

  47. [47]

    Online few-shot gesture learning on a neuromorphic processor.IEEE Jour- nal on Emerging and Selected Topics in Circuits and Systems, 10(4):512–521, 2020

    Kenneth Stewart, Garrick Orchard, Sumit Bam Shrestha, and Emre Neftci. Online few-shot gesture learning on a neuromorphic processor.IEEE Jour- nal on Emerging and Selected Topics in Circuits and Systems, 10(4):512–521, 2020

    work page 2020

  48. [48]

    Dendritic inte- gration: 60 years of progress.Nature neuroscience, 18(12):1713–1721, 2015

    Greg J Stuart and Nelson Spruston. Dendritic inte- gration: 60 years of progress.Nature neuroscience, 18(12):1713–1721, 2015

    work page 2015

  49. [49]

    Efficient rehearsal free zero forgetting continual learning using adaptive weight modulation.arXiv preprint arXiv:2311.15276, 2023

    Yonatan Sverdlov and Shimon Ullman. Efficient rehearsal free zero forgetting continual learning using adaptive weight modulation.arXiv preprint arXiv:2311.15276, 2023

    work page arXiv 2023

  50. [50]

    Measuring robustness to natural distri- bution shifts in image classification.Advances in Neural Information Processing Systems, 33:18583– 18599, 2020

    Rohan Taori, Achal Dave, Vaishaal Shankar, Nicholas Carlini, Benjamin Recht, and Ludwig Schmidt. Measuring robustness to natural distri- bution shifts in image classification.Advances in Neural Information Processing Systems, 33:18583– 18599, 2020

    work page 2020

  51. [51]

    Temporal quality degradation in ai models.Scientific Reports, 12(1):11654, 2022

    Daniel Vela, Andrew Sharp, Richard Zhang, Trang Nguyen, An Hoang, and Oleg S Pianykh. Temporal quality degradation in ai models.Scientific Reports, 12(1):11654, 2022

    work page 2022

  52. [52]

    Rehearsal revealed: The limits and merits of revisiting samples in continual learning

    Eli Verwimp, Matthias De Lange, and Tinne Tuyte- laars. Rehearsal revealed: The limits and merits of revisiting samples in continual learning. InPro- ceedings of the IEEE/CVF International Confer- ence on Computer Vision, pages 9385–9394, 2021

    work page 2021

  53. [53]

    Wanderlust: Online continual ob- ject detection in the real world

    Jianren Wang, Xin Wang, Yue Shang-Guan, and Abhinav Gupta. Wanderlust: Online continual ob- ject detection in the real world. InProceedings of the IEEE/CVF international conference on com- puter vision, pages 10829–10838, 2021

    work page 2021

  54. [54]

    Concept drift adaptation methods under the deep learning framework: A literature review

    Qiuyan Xiang, Lingling Zi, Xin Cong, and Yan Wang. Concept drift adaptation methods under the deep learning framework: A literature review. Applied Sciences, 13(11):6515, 2023

    work page 2023

  55. [55]

    Spiking neural networks and their applications: A review.Brain sciences, 12(7):863, 2022

    Kashu Yamazaki, Viet-Khoa V o-Ho, Darshan Bul- sara, and Ngan Le. Spiking neural networks and their applications: A review.Brain sciences, 12(7):863, 2022

    work page 2022

  56. [56]

    Recent advances in concept drift adaptation methods for deep learning

    Liheng Yuan, Heng Li, Beihao Xia, Cuiying Gao, Mingyue Liu, Wei Yuan, and Xinge You. Recent advances in concept drift adaptation methods for deep learning. InIJCAI, pages 5654–5661, 2022. 16

    work page 2022