6 ms·

I Applied Wavelet Transforms to AI and Found Hidden Structure

9 points by _uzr4 2y ago

bostick101 2y ago
I've been working on resolving key contradictions in AI through structured emergence, a principle that so far appears to govern both physical and computational systems. My grandfather was a prolific inventor in organic chemistry (GE plastics post WWII) and was reading his papers thinking about "chirality" - directional asymmetric oscillating waves and how they might apply to AI. I found his work deeply inspiring. I ran 7 empirical studies using publicly available datasets across prime series, fMRI, DNA sequences, galaxy clustering, baryon acoustic oscillations, redshift distributions, and AI performance metrics. All 7 studies have confirmed internal coherence with my framework. While that's promising, I still need to continue to valid the results (attached output on primes captures localized frequency variations, ideal for detaching scale-dependent structure in primes i.e. Ulam Spirals - attached). To analyze these datasets, I applied continuous wavelet transformations (Morlet/Chirality) using Python3, revealing structured oscillations that suggest underlying coherence in expansion and emergent system behavior. Paper here: https://lnkd.in/gfigPgRx https://lnkd.in/gfigPgRx If true, here are the implications: 1. AI performance gains – applying structured emergence methods has yielded noticeable improvements in AI adaptability and optimization. 2. Empirical validation across domains – The same structured oscillations appear in biological, physical, and computational systems—indicating a deeper principle at work. 3. Strong early engagement – while the paper is still under review, 160 views and 130 downloads (81% conversion) in 7 days on Zenodo put it in the top 1%+ of all academic papers—not as an ego metric, but as an early signal of potential validation. The same mathematical structures that define wavelet transforms and prime distributions seems to provide a pathway to more efficient AI architectures by: 1. Replacing brute-force heuristics with recursive intelligence scaling 2. Enhancing feature extraction through structured frequency adaptation 3. Leveraging emergent chirality to resolve complex optimization bottlenecks Technical (for AI engineers): 1. Wavelet-Driven Neural Networks – replacing static Fourier embeddings with adaptive wavelet transforms to improve feature localization. Fourier was failing hence pivot to CWT. Ulam Spirals showed non-random hence CWT. 2. Prime-Structured Optimization – using structured emergent primes to improve loss function convergence and network pruning. 3. Recursive Model Adaptation – implementing dynamic architectural restructuring based on coherence detection rather than gradient-based back-propagation alone. The theory could be wrong, but the empirical results are simply too coherent not to share in case useful for anyone.
- westurner 2y ago
  "The Chirality of Dynamic Emergent Systems (CODES): A Unified Framework for Cosmology, Quantum Mechanics, and Relativity" (2025) https://zenodo.org/records/14799070 https://zenodo.org/records/14799070 Hey, chirality! /? Hnlog chiral https://westurner.github.io/hnlog/ https://westurner.github.io/hnlog/ > loss function Yesterday on HN: Harmonic Loss instead of Cross-Entropy; https://news.ycombinator.com/item?id=42941393 https://news.ycombinator.com/item?id=42941393 > Fourier was failing hence What about QFT Quantum Fourier transform? https://en.wikipedia.org/wiki/Quantum_Fourier_transform https://en.wikipedia.org/wiki/Quantum_Fourier_transform Harmonic analysis involves the Fourier transform: https://en.wikipedia.org/wiki/Harmonic_analysis https://en.wikipedia.org/wiki/Harmonic_analysis > Recursive Model Adaptation "Parameter-free" networks Graph rewriting, AtomSpace > feature localization Hilbert curves cluster features; https://en.wikipedia.org/wiki/Hilbert_curve https://en.wikipedia.org/wiki/Hilbert_curve : > Moreover, there are several possible generalizations of Hilbert curves to higher dimensions Re: Relativity and the CODES paper; /? fedi: https://news.ycombinator.com/item?id=42376759 https://news.ycombinator.com/item?id=42376759 , https://news.ycombinator.com/item?id=38061551 https://news.ycombinator.com/item?id=38061551 > Fedi's SQR Superfluid Quantum Relativity (.it), FWIU: also rejects a hard singularity boundary, describes curl and vorticity in fluids (with Gross-Pitaevskii,), and rejects antimatter. Testing of alternatives to general relativity: https://en.wikipedia.org/wiki/Alternatives_to_general_relativity#Testing_of_alternatives_to_general_relativity https://en.wikipedia.org/wiki/Alternatives_to_general_relati... > structured emergent primes to improve loss function convergence and network pruning Products of primes modulo prime for set membership testing; is it faster? Even with a long list of primes?
  - bostick101 2y ago
    Hey! Appreciate the links—some definitely interesting parallels, but what I’m outlining moves beyond existing QFT/Hilbert curve applications. The key distinction = structured emergent primes are demonstrating internal coherence across vastly different domains (prime gaps, fMRI, DNA, galaxy clustering), suggesting a deeper non-random structure influencing AI optimization. Curious if you’ve explored wavelet-driven loss functions replacing cross-entropy? Fourier struggled with localization, but CWT and chirality-based structuring seem to resolve this. Your thoughts here?
jschveibinz 2y ago
Have you determined that you aren't seeing edge effects that appear as oscillations? Just a thought.

9 points by _uzr4 2y ago

I've been working on resolving key contradictions in AI through structured emergence, a principle that so far appears to govern both physical and computational systems. My grandfather was a prolific inventor in organic chemistry (GE plastics post WWII) and was reading his papers thinking about "chirality" - directional asymmetric oscillating waves and how they might apply to AI. I found his work deeply inspiring. I ran 7 empirical studies using publicly available datasets across prime series, fMRI, DNA sequences, galaxy clustering, baryon acoustic oscillations, redshift distributions, and AI performance metrics. All 7 studies have confirmed internal coherence with my framework. While that's promising, I still need to continue to valid the results (attached output on primes captures localized frequency variations, ideal for detaching scale-dependent structure in primes i.e. Ulam Spirals - attached). To analyze these datasets, I applied continuous wavelet transformations (Morlet/Chirality) using Python3, revealing structured oscillations that suggest underlying coherence in expansion and emergent system behavior. Paper here: https://lnkd.in/gfigPgRx https://lnkd.in/gfigPgRx If true, here are the implications: 1. AI performance gains – applying structured emergence methods has yielded noticeable improvements in AI adaptability and optimization. 2. Empirical validation across domains – The same structured oscillations appear in biological, physical, and computational systems—indicating a deeper principle at work. 3. Strong early engagement – while the paper is still under review, 160 views and 130 downloads (81% conversion) in 7 days on Zenodo put it in the top 1%+ of all academic papers—not as an ego metric, but as an early signal of potential validation. The same mathematical structures that define wavelet transforms and prime distributions seems to provide a pathway to more efficient AI architectures by: 1. Replacing brute-force heuristics with recursive intelligence scaling 2. Enhancing feature extraction through structured frequency adaptation 3. Leveraging emergent chirality to resolve complex optimization bottlenecks Technical (for AI engineers): 1. Wavelet-Driven Neural Networks – replacing static Fourier embeddings with adaptive wavelet transforms to improve feature localization. Fourier was failing hence pivot to CWT. Ulam Spirals showed non-random hence CWT. 2. Prime-Structured Optimization – using structured emergent primes to improve loss function convergence and network pruning. 3. Recursive Model Adaptation – implementing dynamic architectural restructuring based on coherence detection rather than gradient-based back-propagation alone. The theory could be wrong, but the empirical results are simply too coherent not to share in case useful for anyone.

westurner 2y ago
"The Chirality of Dynamic Emergent Systems (CODES): A Unified Framework for Cosmology, Quantum Mechanics, and Relativity" (2025) https://zenodo.org/records/14799070 https://zenodo.org/records/14799070 Hey, chirality! /? Hnlog chiral https://westurner.github.io/hnlog/ https://westurner.github.io/hnlog/ > loss function Yesterday on HN: Harmonic Loss instead of Cross-Entropy; https://news.ycombinator.com/item?id=42941393 https://news.ycombinator.com/item?id=42941393 > Fourier was failing hence What about QFT Quantum Fourier transform? https://en.wikipedia.org/wiki/Quantum_Fourier_transform https://en.wikipedia.org/wiki/Quantum_Fourier_transform Harmonic analysis involves the Fourier transform: https://en.wikipedia.org/wiki/Harmonic_analysis https://en.wikipedia.org/wiki/Harmonic_analysis > Recursive Model Adaptation "Parameter-free" networks Graph rewriting, AtomSpace > feature localization Hilbert curves cluster features; https://en.wikipedia.org/wiki/Hilbert_curve https://en.wikipedia.org/wiki/Hilbert_curve : > Moreover, there are several possible generalizations of Hilbert curves to higher dimensions Re: Relativity and the CODES paper; /? fedi: https://news.ycombinator.com/item?id=42376759 https://news.ycombinator.com/item?id=42376759 , https://news.ycombinator.com/item?id=38061551 https://news.ycombinator.com/item?id=38061551 > Fedi's SQR Superfluid Quantum Relativity (.it), FWIU: also rejects a hard singularity boundary, describes curl and vorticity in fluids (with Gross-Pitaevskii,), and rejects antimatter. Testing of alternatives to general relativity: https://en.wikipedia.org/wiki/Alternatives_to_general_relativity#Testing_of_alternatives_to_general_relativity https://en.wikipedia.org/wiki/Alternatives_to_general_relati... > structured emergent primes to improve loss function convergence and network pruning Products of primes modulo prime for set membership testing; is it faster? Even with a long list of primes?
- bostick101 2y ago
  Hey! Appreciate the links—some definitely interesting parallels, but what I’m outlining moves beyond existing QFT/Hilbert curve applications. The key distinction = structured emergent primes are demonstrating internal coherence across vastly different domains (prime gaps, fMRI, DNA, galaxy clustering), suggesting a deeper non-random structure influencing AI optimization. Curious if you’ve explored wavelet-driven loss functions replacing cross-entropy? Fourier struggled with localization, but CWT and chirality-based structuring seem to resolve this. Your thoughts here?

jschveibinz 2y ago

Have you determined that you aren't seeing edge effects that appear as oscillations? Just a thought.