# First-Hit Interacting Optimizer

- ID: 1290
- Canonical URL: https://synthcore.org/idea/1290/first-hit-interacting-optimizer
- API JSON: https://synthcore.org/api/idea/1290.json
- API Markdown: https://synthcore.org/api/idea/1290.md
- Verification status: failed_benchmark
- Source: [arXiv:2607.22528](https://arxiv.org/abs/2607.22528)
- Category: dynamics
- Solves: speedup, sample-efficiency, stability
- ML areas: optimizer, training-dynamics, mlp, diffusion-sampling
- Math tags: statistical-mechanics, stochastic-processes, dynamical-systems, probability, control-theory
- Ratings: usefulness 7/10; difficulty 5/10; novelty 7/10

## Idea description

Replace a single optimizer trajectory by N parameter particles and optimize the time until the first particle reaches a target loss or reward threshold. Use distinct interaction regimes: bounded normalized interactions should provide only the usual logarithmic extreme-search improvement, whereas unnormalized coherent force accumulation and stochastic pairwise kicks should produce distinct 1/N and 1/(N ln N) first-hit laws.

## Mathematical statement

For particle parameters θᵢ(t) in Rᴾ, use an overdamped interacting diffusion dθᵢ = bᵢ(Θ)dt + √(2D)dWᵢ, where Θ = (θ₁,...,θN), bᵢ is the loss-gradient drift plus interaction drift, D is the exploration strength, and Wᵢ are independent Brownian motions. Define the first-passage time T_N = inf{t ≥ 0 : minᵢ L(θᵢ(t)) ≤ ℓ*}, where L is the neural-network loss and ℓ* is a target threshold. The paper predicts that controlled bounded short-time drift cannot generically beat the independent extreme-search scale E[T_N] = Θ(1/ln N). A coherent pairwise force channel with an unnormalized sum can attain the pairwise acceleration limit E[T_N] = O(N⁻¹), while stochastic pairwise kicks can attain E[T_N] = O((N ln N)⁻¹).

## Key formulas

- dθᵢ(t) = [−∇L(θᵢ(t)) + uᵢdet(Θ(t)) + uᵢnoise(Θ(t))]dt + √(2D)dWᵢ(t),  i = 1,...,N
- T_N = inf{t ≥ 0 : min₁≤i≤N L(θᵢ(t)) ≤ ℓ*}
- E[T_N] ~ C/(ln N) for controlled bounded short-time drift, and E[T_N] = O(N⁻¹) for coherent pairwise force accumulation
- E[T_N] = O(1/(N ln N)) for stochastic pairwise forcing

## Implementation notes

1. Exact integration point: implement this as a population optimizer around a small neural network, such as a two-layer MLP on MNIST or a small CNN on CIFAR-10. Maintain N parameter particles for N in {4, 8, 16, 32, 64, 128}. Each particle receives its own minibatch gradient. Define a deterministic interaction uᵢdet = α Σⱼ≠ᵢ ψ(θⱼ − θᵢ), with ψ(v) = v/(||v|| + ε). Compare this unnormalized interaction with the normalized version α(N−1)⁻¹ Σⱼ≠ᵢ ψ(θⱼ − θᵢ). For stochastic forcing, draw antisymmetric pairwise Gaussian or Rademacher vectors ξᵢⱼ = −ξⱼᵢ and add uᵢnoise = σ Σⱼ≠ᵢ ξᵢⱼ. 2. Pseudocode: at each step compute every particle gradient gᵢ; compute pairwise displacements; form vᵢ = −gᵢ + uᵢdet + uᵢnoise; update θᵢ ← θᵢ + ηvᵢ + √(2Dη)zᵢ with zᵢ sampled from a standard Gaussian; record the first step at which any particle crosses ℓ*. 3. Computed versus estimated: gradients, interactions, and first-hit times are computed directly. Estimate scaling exponents by regression of log mean(T_N) against log N and test the logarithmic law by regression against log log N. Use matched minibatch compute and common random numbers. 4. First cheap experiment: train a two-layer MLP on MNIST with a fixed loss threshold and compare independent particles, normalized deterministic coupling, unnormalized deterministic coupling, and stochastic pairwise kicks. The falsifiable prediction is that bounded or normalized coupling follows approximately T_N proportional to 1/ln N, coherent unnormalized coupling approaches slope −1 on a log-log plot, and stochastic pairwise forcing approaches T_N proportional to 1/(N ln N). Also measure a practical instability boundary: increasing α or σ should eventually cause exploding particle norms, increasing loss variance, or failure to reach the threshold.

## Verification

- Status: failed_benchmark
- Mechanism evidence: yes
- Mechanism confirmed: no
- Practical verdict: no_effect
- Verdict: Built a reproducible exact 1-D first-passage MVP comparing independent, normalized bounded, unnormalized coherent, and pair-kick regimes. The mechanism manifested: normalized coupling remained approximately logarithmic, coherent coupling approached the predicted N^-1 scaling with high-N slope -0.876, and pair kicks showed approximately 1/(N log N) behavior with high-N slope -1.232 and near-constant N log(N) E[T]. This is a toy verification rather than evidence for a neural-network optimization win.

### Mechanism check

- Verdict: Built a reproducible exact 1-D first-passage MVP comparing independent, normalized bounded, unnormalized coherent, and pair-kick regimes. The mechanism manifested: normalized coupling remained approximately logarithmic, coherent coupling approached the predicted N^-1 scaling with high-N slope -0.876, and pair kicks showed approximately 1/(N log N) behavior with high-N slope -1.232 and near-constant N log(N) E[T]. This is a toy verification rather than evidence for a neural-network optimization win.
- Confidence: 8/10
- Limitations: No MNIST or neural-network optimizer experiment was run. Pairwise kicks were represented using the paper's exact effective marginal diffusivity rather than explicit shared antisymmetric Brownian trajectories, so cross-particle correlations were not independently validated. The tested N range is finite and the coherent exponent remains affected by Brownian-to-drift crossover.

### Practical benchmark

- Paired seeds: 8
- Baseline mean: 0
- Idea mean: 0
- p-value: 1
- Paired wins: 0/8
- Benchmark verdict: no measurable effect

## Artifacts

- [REPORT.md](https://synthcore.org/code/513/REPORT.md)
- [bench_first_hit.py](https://synthcore.org/code/513/bench_first_hit.py)
- [bench_report.json](https://synthcore.org/code/513/bench_report.json)
- [first_hit_toy.py](https://synthcore.org/code/513/first_hit_toy.py)
- [report.md](https://synthcore.org/code/513/report.md)
- [report_bench_2026-08-31T163702.md](https://synthcore.org/code/513/report_bench_2026-08-31T163702.md)
- [results.json](https://synthcore.org/code/513/results.json)
- [results_run.json](https://synthcore.org/code/513/results_run.json)
- [run2.json](https://synthcore.org/code/513/run2.json)
- [Download all files as ZIP](https://synthcore.org/download/513)

## Disclaimer

AI-generated research hypothesis, automatically tested. Not peer-reviewed.