1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
|
# SDIL novelty audit
This file records the closest-prior-art boundary as of 2026-07-22. It is a claim constraint, not
a literature-survey completeness claim. New evidence may narrow the allowed claims further.
## Exact overlap with learned node-perturbation feedback
Lansdell, Prakash, and Kording, [Learning to solve the credit assignment
problem](https://arxiv.org/abs/1906.00889) (ICLR 2020), already train synthetic feedback matrices
from node-perturbation estimates. Their direct-feedback form maps the output error to every hidden
layer and was evaluated in a CIFAR convolutional network. Their public implementation is
[`benlansdell/synthfeedback`](https://github.com/benlansdell/synthfeedback).
With no ordinary apical traffic and `P=0`, SDIL reduces exactly to this idea (up to sign and tensor
conventions):
```text
c = output error
r_l = A_l c
q_l = node-perturbation estimate of -dL/dh_l
A_l <- A_l + eta_A (q_l - A_l c) c^T
```
Independent noise at multiple hidden layers during a shared noisy forward pass is also present in
the earlier method. SDIL's antithetic Rademacher directions, vectorized direction batching, and
explicit cost accounting are useful estimator/implementation choices, but are not the core
algorithmic novelty. The existing no-traffic depth results therefore establish the strength of a
learned-node-perturbation backbone; by themselves they do not establish a Harnett-specific
contribution.
The closest exact baseline is consequently the same implementation with `traffic_mode=none`,
`learn_P=0`, and `use_residual=0`. It should be named **learned direct feedback by node
perturbation (Lansdell et al.)**, not SDIL, in comparisons intended to isolate novelty.
## Overlap with dendritic credit assignment
Apical/basal segregation, neuron-specific dendritic teaching signals, and local three-factor or
burst-dependent plasticity substantially predate SDIL. In addition, Greedy et al.,
[Cell-type-specific cortical feedback coordinates hierarchical credit
assignment](https://www.biorxiv.org/content/10.64898/2026.06.16.732595v1) (BurstCCN, 2026),
explicitly connects a scalable dendritic credit-assignment model to the Francioni/Harnett data.
BurstCCN already:
- models deviations from balanced dendritic feedback as neuron-specific error signals;
- reproduces the opposite residual signs of the P+ and P- BCI populations and their disruption by
NDNF activation;
- trains convolutional models on CIFAR-10 and ImageNet with plastic feedback; and
- provides an [author implementation](https://github.com/neuralml/BurstCCN-journal).
Thus neither "apical dendrites carry vector errors", "the model reproduces Harnett Fig. 5", nor
"a dendritic local rule scales to ImageNet" is a sufficient novelty claim.
Likewise, recursive spatial feedback, learned feedback convolutions, and local
ReLU-gated error propagation are not new by themselves. The post-failure
hierarchical oracle deliberately reconstructs a BP-like local adjoint and may
guide engineering, but it cannot be presented as an SDIL contribution. Any
trainable version must be compared directly with hierarchical FA, learned
feedback alignment/weight-mirroring methods, and BurstCCN. The Harnett-specific
claim remains the neutral-period somato-dendritic innovation operation under
mixed traffic.
## Candidate SDIL contribution
The candidate contribution is the combination
```text
a_l = A_l c + t_l mixed apical traffic
P_l(h_l) ~= E[t_l | h_l, neutral] per-cell neutral predictor
r_l = a_l - P_l(h_l) somato-dendritic innovation
Delta W_l proportional to r_l * eligibility_l
```
where the subtractive per-cell innovation, rather than raw apical activity, is the teaching
variable. The feedback vectorizer `A_l` may be learned using Lansdell-style causal perturbations;
that component is inherited, not claimed as new. The new empirical question is whether
residualization is load-bearing when teaching signals share an apical channel with ordinary
somatic/contextual feedback.
Neutral-period fitting is an algorithmic adaptation rather than a direct statement of the Harnett
paper. It addresses an identifiability problem: task-period regression can subtract the predictable
part of `A_l c` whenever output error correlates with somatic state, while neutral-period regression
identifies the traffic relationship without observing the teaching component.
The formal justification is an application of the standard conditional-expectation projection
identity, not a new mathematical theorem. Among all predictors measurable from the chosen somatic
statistic, the neutral conditional mean uniquely minimizes residual nuisance power and leaves an
innovation orthogonal to every such predictor. The implemented per-cell affine model realizes the
restricted projection onto `{1,h_i}`. Positive norm matching cannot change the raw signal's cosine
direction at a fixed state, so the matched-raw ablation specifically distinguishes subtraction from
gain control. `THEORY.md` states the assumptions and the executable finite-sample identities.
## Claims currently allowed
- The inherited learned-feedback backbone preserves performance with depth better than fixed DFA
in the audited flattened-CIFAR MLP setting.
- Frozen controls show that per-cell residualization protects learning from soma-coupled traffic.
It mitigates but does not fully remove endogenous top-down traffic, so arbitrary contextual
innovation is not an allowed teaching-signal claim.
- Neutral residualization is the minimum-power nuisance removal within the predictor's somatic
function class. For the implemented diagonal affine class this is a narrow, testable statement,
not a claim of optimal population-level denoising.
- Simultaneous batched perturbations have a frozen hardware-independent advantage: K1/every4
retains 112.9% of the K16/every4 gain over DFA with 16x fewer logical loss queries, 11x less
calibration work, and 5.3x less total forward-equivalent work on CIFAR-10 d20/w64.
- Direct node perturbation solves the frozen useful-depth task while the amortized context
vectorizer does not. This identifies feedback amortization as the current engineering
bottleneck; it is not a novelty claim and the direct estimator's 68.4x work precludes presenting
it as the scalable algorithm.
## Claims currently forbidden
- Node perturbation training of `A_l` is novel.
- Simultaneously perturbing all hidden layers is novel.
- The no-traffic scaling panel demonstrates the necessity of somato-dendritic residuals.
- SDIL is the first dendritic method to scale to deep vision tasks.
- Reproducing the already-published Harnett/BurstCCN qualitative signatures is an oral-level
biological contribution.
## Evidence that can restore a strong paper
1. Show across independently generated, endogenous traffic families that residualization beats
raw and norm-matched raw feedback, while the no-traffic methods coincide.
2. Compare directly with Lansdell-style learned feedback and BurstCCN under validation-selected,
disclosed compute/query budgets.
3. The conditional-projection/SNR result and task-period absorption prediction are now proved and
executable. A broader paper would still benefit from a separately frozen failure test where
traffic depends on the teaching signal after conditioning on the available soma statistic.
4. For a biological contribution, make and test a prediction not used by Harnett or BurstCCN. Raw
event-level data and analysis code require author coordination; the paper exposes only plotted
source data and states that full data/code are available on request.
5. For a scaling contribution, target standard architectures and a simplicity/cost regime not
already occupied by BurstCCN, rather than treating ImageNet alone as novelty.
|
