A Spatiotemporal Extension of the Neuromorphic DBSCAN Implementation
Charles P. Rizzo, James S. Plank
cs.NE
Quick verdict
Extends neuromorphic DBSCAN SNNs with causal temporal look-back via memory-neuron feedback; precise resource math and open code, but no efficacy results.
Summary
The paper extends the authors' prior neuromorphic DBSCAN implementations - the throughput-oriented 'flat' and the smaller 'systolic' spiking constructions - to operate spatiotemporally on event-camera streams. New Input_Mem and Core_Mem neuron collections buffer the previous eps_t frames' inputs and core classifications (via synaptic delays, e.g., C+2eps+4 timesteps in the systolic variant) and recurrently feed them back, so prior activity contributes to current neighborhoods. Unlike Zhang et al.'s centered 3D voxel neighborhood, the design uses a causal look-back window, classifying events immediately without waiting for future frames. The paper derives closed-form neuron/synapse counts (temporal flat: RC(5+2eps_t) neurons; temporal systolic: R((4eps+5)+2eps_t(2eps+1))), reports constructed network sizes for 10x10 and 260x346 resolutions (up to 1,169,480 neurons / 72,770,360 synapses flat versus 24,180 / 232,460 systolic), shows multi-edge support could replace memory neurons with delayed synapses, and gives partial sub-region implementations with rescaled delays for constrained hardware. All constructions are open-source. Key limitation: the work is purely architectural - no experiments on event data, no efficacy comparison against spatial-only or competing denoisers, and no hardware latency/throughput measurements; efficacy evaluation is explicitly deferred to future work.
Review complete
Assessment at a glance
4
risk flags identified
● high 1
● medium 2
● low 1
4 strengths · 4 weaknesses · 5 author questions
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Adding eps_t blocks of Input_Mem and Core_Mem neurons with delay-based recurrent feedback extends both the flat and systolic neuromorphic DBSCAN constructions to spatiotemporal operation on event streams.
- The causal look-back window classifies events immediately, avoiding the at-least-eps_t-frame classification latency inherent to centered-window spatiotemporal DBSCAN such as Zhang et al.'s.
- The temporal networks are precisely characterized in size and runtime: RC(5+2eps_t) neurons for temporal flat, R((4eps+5)+2eps_t(2eps+1)) for temporal systolic, reaching 72.8M synapses (flat) and 232,460 synapses (systolic) at Davis346 resolution with eps=eps_t=4.
- If the neuromorphic processor supported multi-edges, the memory neurons could be replaced by 2eps_t*RC*N additional delayed synapses in the flat construction.
- Partial sub-region implementations extend the time-for-space trade to the temporal variant by setting the feedback delays equal to the sub-region schedule period (e.g., 910 timesteps for 260x346 with 10x10 kernels).
Strengths
Causal, low-latency temporal design
Section III / Fig. 4 replaces Zhang et al.'s centered temporal window (which incurs at least eps_t frames of forward-looking latency) with a look-back window so events classify immediately - a well-motivated distinction for live event-camera pipelines.
Verifiable resource characterization
Tables II-III give closed-form neuron/synapse totals that check out on inspection (e.g., temporal systolic neurons R((4eps+5)+2eps_t(2eps+1)) = 24,180 for 260x346 with eps=eps_t=4, exactly as in Table IV), with timesteps-to-solution and timesteps-to-reuse cleanly separated.
Honest hardware pragmatism
Open-source constructions
Weaknesses
No empirical evaluation at all
No hardware deployment despite edge-deployment motivation
Constructed totals deviate from stated formulas without explanation
Incomplete treatment where it matters most
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
- All 4 strengths & 4 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What does this paper add over the authors' earlier neuromorphic DBSCAN work?
It adds a time dimension: extra memory neuron collections (Input_Mem, Core_Mem) store the events and core classifications of the previous eps_t frames and feed them back through delayed synapses, so clustering and denoising decisions consider recent history instead of only the current frame, matching how moving objects leave temporally correlated events.
Is the code for the spatiotemporal neuromorphic DBSCAN networks available?
Yes. The network constructions are open source at github.com/TENNLab-UTK/dbscan, the same repository as the authors' earlier spatial-only implementation.
How is this different from Zhang et al.'s 3D spatiotemporal DBSCAN denoising for event cameras?
Zhang et al. use a temporal window centered on the event being classified, so the system must wait eps_t future frames before deciding (added latency). This paper's design looks only backward in time, classifying each event immediately; the paper argues this suits live deployment but reports no accuracy comparison between the two.
Does the spatiotemporal version actually denoise or cluster better than the spatial-only version?
This paper does not say - it contains no experiments. The authors state that comparing efficacy between the spatiotemporal and earlier spatial networks is future work, along with real hardware deployments.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1-5 could not be visually analyzed (text extraction garbles diagram content); connectivity claims rest on the prose and Tables I-III.
- The base flat/systolic constructions are formally defined in prior work [21], which is not included in the extracted text; this review assesses the extension as summarized here.
- The paper has no results section, so 'reproReadiness' reflects the reproducibility of the network constructions, not replication of empirical findings.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.