Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

PhasorFlow: A Python Library for Unit Circle Based Computing

T0 review · 4 major / 6 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read PhasorFlow claims that computing on the unit circle with unitary interference gates is a deterministic, lightweight alternative to neural networks and quantum circuits, backed by a 16-parameter classifier and a Fourier-mixing transformer.

desk verdict The library idea and the unitary math are fine, but Table 4 cannot have come from the stated single-layer VPC, so the paper as written is internally inconsistent and needs a rerun before it deserves referee time. read the letter →

arxiv 2603.15886 v4 pith:TXAKLI22 submitted 2026-03-16 cs.LG cs.AI

Dibakar Sigdel , Namuna Panday This is my paper · ORCID
classification cs.LGcs.AI
keywords unitcirclecomputingphasorcircuitsvariationaldiscreteFouriertransformtransformerclassificationneuromorphicphasecoherence
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish the unit circle S^1 as a viable computational manifold, where each datum becomes a complex phasor e^{iφ}. It formalizes Phasor Circuits, a gate algebra of rotations, pairwise mixing, and global Fourier transforms, and shows two trainable descendants work: a Variational Phasor Circuit that classifies a non-linear synthetic boundary with 100% validation accuracy using only 16 phase parameters, and a Phasor Transformer whose attention is replaced by a parameter-free DFT mixer. If true, phase-aware tasks—oscillatory signals, financial volatility, associative memory—could be handled deterministically on ordinary hardware with far fewer trainable parameters than dense neural networks. The paper also claims its phase-coherence observable works as an unsupervised anomaly detector.

What carries the argument

Phasor circuit state z=(e^{iφ_1},…,e^{iφ_N})∈T^N; gates: Shift S_k(θ)=diag(…,e^{iθ},…), Mix M=(1/√2)[[1,i],[i,1]], and DFT F_N. The Phasor Transformer block B(θ)=S(θ_post)F_T S(θ_pre) is the key identity: diagonal phase projections sandwich a parameter-free discrete Fourier transform as the token mixer, cutting mixing cost to O(T log T) and keeping parameters linear in context length. Phase coherence C=(1/N)|Σ z_k| serves as a parameter-free observable for anomaly detection.

What would settle it

Run the Phasor Transformer on a non-Fourier sequence task where local or causal structure matters (e.g., a tiny language model or sequence-copying). If its accuracy is far worse than a same-size linear-attention baseline, the claim that the DFT mixer suffices as a general attention replacement fails. Alternatively, construct a phase-encoded classification dataset whose boundary is not expressible by a sum of cosines; if a 16-parameter VPC cannot exceed chance where a small MLP can, the representational-capacity claim is falsified.

Watch

Extended reading notes

Core claim

Representing data as phases on the N-torus and evolving the state through unitary gates—diagonal phase shifts, beam-splitter mixes, and a global DFT—yields a deterministic, classically executable paradigm. The central demonstrations: a single-layer Variational Phasor Circuit with 16 trainable parameters reaches 100% validation accuracy on a synthetic non-linear boundary, and a two-block Phasor Transformer (about 128 parameters) regresses a noisy sum-of-sines sequence with MSE ≈ 0.07 while its DFT mixer has zero trainable parameters. The paper argues this makes unit-circle computing a principled alternative to neural networks and quantum circuits for phase-structured problems.

Load-bearing premise

The load-bearing premise is that a fixed, parameter-free DFT token-mixing layer can capture enough token interaction to replace learned self-attention—yet the only sequence benchmark is a Fourier-friendly sum-of-sines forecast, with no language, vision, or general sequence validation.

Editorial extensions

If this is right

  • If the Phasor Circuit formalism holds, phase-structured machine learning tasks can run deterministically on classical hardware without quantum sampling overhead.
  • A 16-parameter VPC reaching 100% on a non-linear boundary indicates extreme parameter compression is possible when the decision structure is phase-based.
  • The Phasor Transformer's DFT mixer gives O(T log T) token mixing with zero parameters in the mixer, reducing sequence-modeling cost dramatically if it generalizes.
  • Phase coherence, being unsupervised and parameter-free, offers a ready cross-domain anomaly indicator such as financial regime shifts.
  • Unit-circle gates natively implement associative memory, neural binding, and even algorithmic logic, unifying deep learning with neuromorphic simulation in one library.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Testable extension we would propose: stress the DFT-mixer claim on a non-Fourier sequence task like a small language model or sequence-copying; collapse there would falsify the general attention-replacement premise.
  • The 'alternative to quantum circuits' claim is best read as an alternative to variational quantum circuits, not to algorithms like Shor's with exponential Hilbert-space speedups, since the phasor state space is N-dimensional.
  • Because the appendix shows the normalization Jacobian becomes singular near destructive interference, we infer that phase-based architectures enforcing strict unit-modulus pull-backs will face gradient pathologies in exactly those regions.
  • The coherence-drop detector may transfer to seizure detection or industrial fault monitoring, since it relies only on multivariate phase alignment rather than learned features.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper introduces PhasorFlow, an open-source Python library for computing with unit-modulus complex numbers (phasors). It formalizes a Phasor Circuit model with 22 gate primitives, proposes a Variational Phasor Circuit (VPC) classifier analogous to variational quantum circuits, and introduces a Phasor Transformer that replaces self-attention with a parameter-free DFT token-mixing layer. Validation is carried out on synthetic tasks: non-linear binary classification, autoregressive time-series forecasting, financial volatility detection, associative memory, neural binding, and period finding. The authors claim that the results establish unit-circle computing as a deterministic, lightweight, and mathematically principled alternative to classical neural networks and quantum circuits.

Significance. The paper's elementary unitary mathematics (phase shifts, Mix gates, DFT) is correct and the library appears to provide a clean, deterministic, open-source implementation of a phase-based circuit model. The idea of a parameter-free DFT token mixer is an interesting adaptation of FNet to a unitary-circuit setting. If the empirical claims were reliable, the parameter-efficiency results would be noteworthy. However, the central quantitative claims are not supported by the reported experiments: the flagship VPC result is internally inconsistent with the stated architecture, and the remaining results are overstated by large margins. The paper does not currently provide credible evidence for the advertised paradigm.

major comments (4)
  1. [§3.2.1 and §6.1 (Eqs. 34, 37, 42; Table 4)] The single-layer VPC with N=16 and |θ|=16 is defined by V(θ)=U_local ∏ S_k(θ_k) with U_local=∏_{k even} M_{k,k+1}. The pair-wise mixers act on disjoint pairs (0,1), (2,3), …, so thread 0 after the circuit depends only on ϕ0, ϕ1 and θ0, θ1: z_f,0 = (e^{i(θ0+ϕ0)} + i e^{i(θ1+ϕ1)})/√2. The stated label y=1[Σ_{k=1}^{16} cos(ϕ_k)>0] depends on all 16 phases. The readout of thread 0 cannot achieve 100% accuracy — the Bayes-optimal accuracy using only ϕ0 and ϕ1 is far below 100%. Table 4's '100.0%' is also contradicted by Fig. 9D, which reports 94.5% accuracy (189/200). The headline result is internally inconsistent with the architecture as described.
  2. [§6.2, Table 5] The sequence regression benchmark reports test MSE≈0.075 for the Phasor Transformer versus ≈0.003 for a standard transformer — a 25-fold increase in error — and yet the text describes the result as 'nearly identical test set interpolation' and 'a highly viable alternative.' Moreover, the only sequence task is a sum of three sinusoids with additive Gaussian noise, which is already Fourier-friendly and does not test whether a fixed DFT mixer can substitute for self-attention on general sequence modeling. The abstract's claim that PhasorFlow is 'a deterministic, lightweight, and mathematically principled alternative to classical neural networks' is not supported by this evidence.
  3. [§6.3, Fig. 11] The financial volatility detector is described as yielding 'strong crisis/non-crisis discrimination,' yet the reported confusion matrix gives F1=0.386, precision=0.338, recall=0.450. The detection threshold 0.347 appears to be selected on the same synthetic series used to compute the confusion matrix; no train/test split, cross-validation, or sensitivity analysis is reported. An F1 of 0.386 is not strong, and the post hoc threshold selection raises concerns that the result is overfitted to one synthetic path.
  4. [§6.4, Fig. 12D] The associative-memory capacity law '0.14N' is presented as a finding, but the line '0.14N=2.2' in Fig. 12D is fit to the recall curve of the very same experiment. No independent evaluation across different N, different storage patterns, or different noise levels is provided; no confidence intervals or capacity definition are given. This is a post hoc fit, not a validated property of the model, and the caption's claim that capacity 'remains robust' is unsupported by the reported data.
minor comments (6)
  1. [§8 vs. §6.1] The conclusion states 'as few as 32 trainable phase parameters,' while Section 6.1 and Table 4 report |θ|=16. Please reconcile.
  2. [§6.1, Fig. 9] The text says 'optimized using Adam Backpropagation over 200 epochs,' but Fig. 9B/C show x-axes only to 40 epochs. Clarify whether the figure shows a subset or a different run.
  3. [§3.2.1, Eq. (42)] The probability map is written p=sin(ϕ0)+1/2, which ranges in [−0.5,1.5]; it should be (sin(ϕ0)+1)/2 to lie in [0,1].
  4. [§4.1] The models module is said to offer a 'PhasorGAN', but no PhasorGAN is defined, evaluated, or referenced elsewhere in the manuscript.
  5. [§3.3.1, Eq. (46)] The encoding ϕ_t = (s_t / max|s|)·π/2 divides by max|s|; the case max|s|=0 (e.g., a constant zero input) is not handled.
  6. [§5.2 vs. §6.2] Section 5.2 states a context length T=32 and a 2-block architecture with 128 parameters; Section 6.2 uses T=10 and reports 50 parameters. The discrepancy is not explained.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the core phasor algebra and DFT token mixer are defined independently of the reported results; the main weaknesses are internal inconsistencies and synthetic-only validation, not circular reductions.

full rationale

The derivation chain is not circular. The VPC (Eqs. 34-43) is a genuinely parameterized map from encoded phases to a readout; the 100% validation claim in Table 4 is inconsistent with the stated single-layer circuit because the readout phase depends only on threads 0 and 1 (from Eq. 37's pairwise mixers applied in Eq. 38), while the label (Eq. 54) is a function of all 16 phases. That is a reproducibility/correctness problem, not a self-referential reduction: the claimed result is impossible for the defined model, not identical to it by construction. The financial threshold in Fig. 11 (0.347) is presented without a selection procedure, so one cannot demonstrate that it was fit to the same series; no quoted equation reduces the detection result to that threshold. The '0.14N' memory capacity in Fig. 12D is explicitly labeled empirical, i.e., a fit to its own recall curve, and is not used to derive a further prediction. Section 7.5's references to unpublished 'Phasor-to-Music' and 'Phasor-to-Qubit' proofs of concept are future directions, not load-bearing evidence. Section 7.4 concedes 'The applications in this work use synthetic datasets,' which is a limitation rather than circularity. The only self-citation (ref [7], Sigdel et al. 2023) appears in background motivation (Section 1) and is not load-bearing. Hence no circular step is established; at most a low-score note for a minor non-load-bearing self-citation and several evidence gaps.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No new physical entities are introduced; 'phasor threads', VPCs, and the Phasor Transformer block are software abstractions, not independent postulates with falsifiable handles. The main ledger burden is in fitted parameters: model rotations, the financial threshold, and the capacity curve.

free parameters (4)
  • VPC phase parameters θ_k (N per layer) = N=16; trained with Adam lr=0.1 for 200 epochs on synthetic classification
    Eq. (36): trainable shifts are the entire model; final validation accuracy is 100% on the authors' synthetic sum-of-cosines rule.
  • Phasor Transformer phase parameters θ_pre, θ_post (2T per block) = T=10, 2 blocks, 50 trainable parameters
    Eq. (52): trainable phase rotations around the fixed DFT mixer; reported test MSE ≈ 0.075 on the synthetic sinusoid forecast.
  • Financial coherence threshold = 0.347
    Fig 11B: threshold used to turn the coherence time series into crisis/non-crisis labels; it appears selected post hoc on the same 300-day synthetic series used for the confusion matrix (F1=0.386).
  • Associative memory capacity coefficient = 0.14N (with N=2.2 in the figure)
    Fig 12D: a fitted line '0.14N=2.2' is drawn through the empirical recall-vs-stored-pattern-count curve; no derivation or independent test is given.
assumptions (5)
  • standard math Unitary matrices preserve ℓ2 norm; the DFT matrix is unitary
    Used in Theorem 2.1 and the DFT gate section; standard linear algebra.
  • domain assumption Synthetic tasks are representative of real phase-structured problems
    Sections 5-6 use only synthetic data; Section 7.4 concedes: 'Validation on massive real-world complex continuous datasets ... would strengthen the empirical claims.'
  • ad hoc to paper A fixed DFT token mixer can substitute for self-attention on the considered tasks
    Borrowed from FNet [11], but the only sequence benchmark is a sum of three sinusoids, a signal that is already Fourier-friendly; no language or general sequence test is run.
  • domain assumption Phase coherence C(t) is a valid unsupervised volatility indicator
    Eq. (14) and Section 5.3; the threshold 0.347 is chosen empirically and no out-of-sample test is provided.
  • ad hoc to paper Associative memory capacity follows the fitted 0.14N curve
    Fig 12D fits a capacity law to its own data without derivation or comparison to known Hopfield capacity bounds.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PhasorFlow: A Python Library for Unit Circle Based Computing." pith.science (2026). https://pith.science/paper/TXAKLI22

@misc{pith2026260315886,
  author       = {Pith},
  title        = {Pith review of: PhasorFlow: A Python Library for Unit Circle Based Computing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TXAKLI22}},
  note         = {Machine review of arXiv:2603.15886}
}
abstract

We present PhasorFlow, an open-source Python library for computing on the $S^1$ unit circle. Inputs are encoded as complex phasors $z=e^{i\phi}$ on the $N$-torus ($\mathbb{T}^N$); as computation proceeds through unitary wave-interference gates, global norm is preserved while components drift into $\mathbb{C}^N$, letting algorithms leverage continuous geometric gradients. PhasorFlow makes three contributions. First, we formalize the Phasor Circuit model ($N$ threads, $M$ gates) with a 22-gate library spanning standard-unitary, non-linear, neuromorphic, and encoding operations under full matrix-algebra simulation. Second, we introduce the Variational Phasor Circuit (VPC), a trainable phase-native classifier analogous to variational quantum circuits. Third, we introduce the Phasor Transformer block and Large Phasor Model (LPM), replacing $QK^TV$ attention with a parameter-free DFT token-mixing layer. We validate the framework on financial volatility detection, neuromorphic associative memory, neural binding, period finding, and algorithmic logic applications that are unique to the library. This positions unit-circle computing as a deterministic, lightweight paradigm on classical hardware. Available at https://github.com/mindverse-computing/phasorflow.

Figures

Figures reproduced from arXiv: 2603.15886 by the authors.

Figure 1
Figure 1. The three paradigms of computation. PhasorFlow introduces the Unit Circle [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The geometric evolution of the internal phase manifold strictly depends connecting [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. An example of an N = 5 continuous Phasor Circuit constructed from parameterized Shift (S) gates and fixed entangling Mix (M) gates, visually analogous to a parameterized quantum circuit cascade. (k, k) and ones elsewhere: Sk(θ) = diag(1, . . . , eiθ |{z} k-th , . . . , 1). (15) Since |e iθ| = 1, the diagonal matrix Sk(θ) is trivially unitary: Sk(θ)Sk(θ) † = I. Applied to a single phasor state e iϕ, the 1 × 1 matrix … view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Circuit representation of the Shift gate acting on a single computation thread. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Circuit representation of the Mix gate, entangling two adjacent continuous phase [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Circuit representation of the N = 5 Discrete Fourier Transform (DFT) gate. Unlike strictly local or pairwise operations, the DFT acts intrinsically as an all-to-all global unitary operator, extracting Fourier-basis classical wave interference phases across the entire t…
Figure 7
Figure 7. Figure 7: The continuous physical circuit architecture of a Variational Phasor Circuit (VPC). Exter [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: The continuous physical circuit architecture of the Phasor Transformer Block [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: VPC binary classification performance on the synthetic dataset, illustrating the loss [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: Phasor Transformer performance on sequence benchmarking, detailing the learning [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]
Figure 11
Figure 11. Figure 11: Financial volatility detection with phasor phase coherence. The indicator drops [PITH_FULL_IMAGE:figures/full_fig_p024_11.png]
Figure 12
Figure 12. Figure 12: Associative memory recovery and scaling. Corrupted binary phase patterns converge [PITH_FULL_IMAGE:figures/full_fig_p025_12.png]
Figure 13
Figure 13. Figure 13: Neural binding through phase synchronization. LIP local coupling rapidly collapses [PITH_FULL_IMAGE:figures/full_fig_p026_13.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. The Phasor Transformer: Resolving Attention Bottlenecks on the Unit Circle

    cs.LG 2026-03 conditional novelty 4.5 of 10

    A DFT-plus-phase-shift “Phasor Transformer” stacks into an LPM that forecasts synthetic multi-frequency series with O(N log N) mixing and far fewer parameters than self-attention, but higher error.

Reference graph

Works this paper leans on

28 extracted references · 1 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Cambridge University Press, 2010

    Michael A Nielsen and Isaac L Chuang.Quantum Computation and Quantum Information: 10th Anniversary Edition. Cambridge University Press, 2010

    2010

  2. [2]

    Springer, 2012

    Akira Hirose.Complex-Valued Neural Networks: Theories and Applications. Springer, 2012

    2012

  3. [3]

    On the importance of phase in neural representations.Proceedings of the National Academy of Sciences, 113(42):11808–11813, 2016

    Mark Tygert, Emma Bruner, Vladimir Rokhlin, and Arthur Szlam. On the importance of phase in neural representations.Proceedings of the National Academy of Sciences, 113(42):11808–11813, 2016

    2016

  4. [4]

    Qiskit: An open-source framework for quantum computing

    Qiskit Contributors. Qiskit: An open-source framework for quantum computing. 2024

    2024

  5. [5]

    Oxford University Press, 2006

    György Buzsáki.Rhythms of the Brain. Oxford University Press, 2006

    2006

  6. [6]

    Cambridge university press, 1996

    Albert Goldbeter.Biochemical oscillations and cellular rhythms: the molecular bases of periodic and chaotic behaviour. Cambridge university press, 1996

    1996

  7. [7]

    Oxidative stress and cellular resonances.MDPI Molecular Sciences, 2023

    Dibakar Sigdel et al. Oxidative stress and cellular resonances.MDPI Molecular Sciences, 2023

    2023

  8. [8]

    Market agentome: Multi-agent phase dynamics in trading

    Quantumlink Capital Holding. Market agentome: Multi-agent phase dynamics in trading. Technical report, Quantumlink Capital Holding, 2025

    2025

Show all 28 references
  1. [9]

    Parameterized quan- tum circuits as machine learning models.Quantum Science and Technology, 4(4):043001, 2019

    Marcello Benedetti, Erika Lloyd, Stefan Sack, and Mattia Fiorentini. Parameterized quan- tum circuits as machine learning models.Quantum Science and Technology, 4(4):043001, 2019

    2019

  2. [10]

    Variational quantum algorithms.Nature Reviews Physics, 3(9):625–644, 2021

    Marco Cerezo, Andrew Arrasmith, Ryan Babbush, Simon C Benjamin, Suguru Endo, Keisuke Fujii, Jarrod R McClean, Kosuke Mitarai, Xiao Yuan, Lukasz Cincio, and Patrick J Coles. Variational quantum algorithms.Nature Reviews Physics, 3(9):625–644, 2021

    2021

  3. [11]

    James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, and Santiago Ontanon. FNet: Mixing tokens with Fourier transforms.Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4296–4313, 2022. 29

    2022

  4. [12]

    Circuit-centric quantum classifiers.Physical Review A, 101(3):032308, 2020

    Maria Schuld, Alex Bocharov, Krysta M Svore, and Nathan Wiebe. Circuit-centric quantum classifiers.Physical Review A, 101(3):032308, 2020

    2020

  5. [13]

    An algorithm for the machine calculation of complex Fourier series.Mathematics of Computation, 19(90):297–301, 1965

    James W Cooley and John W Tukey. An algorithm for the machine calculation of complex Fourier series.Mathematics of Computation, 19(90):297–301, 1965

    1965

  6. [14]

    Neural networks and physical systems with emergent collective computa- tional abilities.Proceedings of the National Academy of Sciences, 79(8):2554–2558, 1982

    John J Hopfield. Neural networks and physical systems with emergent collective computa- tional abilities.Proceedings of the National Academy of Sciences, 79(8):2554–2558, 1982

    1982

  7. [15]

    Associative memory in a network of tunable oscillators.Biological Cybernetics, 81(4):319–321, 1999

    Frank C Hoppensteadt and Eugene M Izhikevich. Associative memory in a network of tunable oscillators.Biological Cybernetics, 81(4):319–321, 1999

    1999

  8. [16]

    High-capacity memory via phase-coded synapses.Neuron, 66(2):250–264, 2010

    Ila R Fiete, Yoram Burak, and Ted Brookings. High-capacity memory via phase-coded synapses.Neuron, 66(2):250–264, 2010

    2010

  9. [17]

    Hopfield networks is all you need.International Conference on Learning Representa- tions, 2021

    Hubert Ramsauer, Bernhard Schäfl, Johannes Lehner, Philipp Seidl, Michael Widrich, Thomas Adler, Lukas Gruber, Markus Holzleitner, Milena Pavlović, Geir Kjetil Sandve, et al. Hopfield networks is all you need.International Conference on Learning Representa- tions, 2021

    2021

  10. [18]

    Self-entrainment of a population of coupled non-linear oscillators.In- ternational Symposium on Mathematical Problems in Theoretical Physics, pages 420–422, 1975

    Yoshiki Kuramoto. Self-entrainment of a population of coupled non-linear oscillators.In- ternational Symposium on Mathematical Problems in Theoretical Physics, pages 420–422, 1975

    1975

  11. [19]

    Springer Science & Business Media, 2001

    Arthur T Winfree.The geometry of biological time. Springer Science & Business Media, 2001

    2001

  12. [20]

    Holographic reduced representations.IEEE Transactions on Neural Net- works, 6(3):623–641, 1995

    Tony A Plate. Holographic reduced representations.IEEE Transactions on Neural Net- works, 6(3):623–641, 1995

    1995

  13. [21]

    Attention is all you need.Advances in Neural Information Processing Systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in Neural Information Processing Systems, 30, 2017

    2017

  14. [22]

    Algorithms for quantum computation: discrete logarithms and factoring

    Peter W Shor. Algorithms for quantum computation: discrete logarithms and factoring. Proceedings 35th Annual Symposium on Foundations of Computer Science, pages 124–134, 1994

    1994

  15. [23]

    Deep complex networks.International Conference on Learning Representations, 2018

    Chiheb Trabelsi, Olexa Bilaniuk, Ying Zhang, Dmitriy Serdyuk, Sandeep Subramanian, Joao Felipe Santos, Soroush Mehri, Negar Rostamzadeh, Yoshua Bengio, and Christopher J Pal. Deep complex networks.International Conference on Learning Representations, 2018

    2018

  16. [24]

    Unitary evolution recurrent neural networks.Proceedings of the 33rd International Conference on Machine Learning, pages 1120–1128, 2016

    Martin Arjovsky, Amar Shah, and Yoshua Bengio. Unitary evolution recurrent neural networks.Proceedings of the 33rd International Conference on Machine Learning, pages 1120–1128, 2016

    2016

  17. [25]

    Full- capacity unitary recurrent neural networks.Advances in Neural Information Processing Systems, 29, 2016

    Scott Wisdom, Thomas Powers, John Hershey, Jonathan Le Roux, and Les Atlas. Full- capacity unitary recurrent neural networks.Advances in Neural Information Processing Systems, 29, 2016

    2016

  18. [26]

    Computing on functions using randomized vector representations.arXiv preprint arXiv:2109.03429, 2021

    E Paxon Frady, Denis Kleyko, and Friedrich T Sommer. Computing on functions using randomized vector representations.arXiv preprint arXiv:2109.03429, 2021

    2021 arXiv

  19. [27]

    Quaternionic neural networks and rotation invariance.IEEE Transactions on Neural Networks, 14(3):687–693, 2003

    Teijiro Isokawa, Nobuyuki Matsui, and Motoaki Kamiura. Quaternionic neural networks and rotation invariance.IEEE Transactions on Neural Networks, 14(3):687–693, 2003

    2003

  20. [28]

    MIT press, 2007

    Eugene M Izhikevich.Dynamical Systems in Neuroscience: The Geometry of Excitability and Bursting. MIT press, 2007. 30 A Geometry of Interference: Leaving theN-Torus This appendix provides a rigorous mathematical demonstration of how unitary interference op- erations genericall...

    2007

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.