pith. machine review for the scientific record. sign in

arxiv: 2604.06561 · v1 · submitted 2026-04-08 · 📡 eess.IV · cs.LG

Recognition: 2 theorem links

· Lean Theorem

Accelerating 4D Hyperspectral Imaging through Physics-Informed Neural Representation and Adaptive Sampling

Chi-Jui Ho , Harsh Bhakta , Wei Xiong , Nicholas Antipa
Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:41 UTC · model grok-4.3

classification 📡 eess.IV cs.LG
keywords hyperspectral imagingphysics-informed neural representationadaptive sampling2DIR spectroscopy4D reconstructionmultilayer perceptronsparse samplingspectral recovery
0
0 comments X

The pith

A multilayer perceptron reconstructs full 4D spectra from just 1/32 of the usual measurements in 2DIR experiments.

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

The paper introduces a neural representation that learns to map sparsely sampled 4D coordinates directly to spectral intensities. It then fills in the missing dense data while an adaptive routine picks the next most useful points to measure based on reconstruction loss. This setup is tested on experimental 2D infrared spectroscopy data that normally demands exhaustive Nyquist sampling and long signal averaging. The result is faithful recovery of both steady-state and oscillating spectral features at a fraction of the cost. If correct, such experiments could finish in minutes instead of hours, opening the door to time-resolved studies of changing molecular systems.

Core claim

A physics-informed multilayer perceptron models the relationship between sub-sampled 4D coordinates and their spectral intensities, allowing dense 4D spectra to be recovered from limited observations. When combined with loss-aware adaptive sampling, the method achieves high-fidelity reconstruction of both oscillatory and non-oscillatory dynamics using only 1/32 of the full sampling budget, reducing experiment time by up to 32-fold in 2DIR hyperspectral measurements.

What carries the argument

Multilayer perceptron that takes 4D coordinates as input and outputs spectral intensity, trained on sparse experimental points and paired with iterative loss-aware adaptive sampling.

If this is right

  • High-fidelity recovery of both oscillatory and non-oscillatory spectral dynamics from limited samples.
  • Up to 32-fold reduction in total acquisition time for 4D hyperspectral imaging.
  • A scalable approach that applies to any hypercube data collection task in multidimensional spectroscopy.
  • Enables faster chemical imaging of transient biological and material systems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same coordinate-to-intensity model could be tested on other high-dimensional modalities where acquisition time is the main bottleneck, such as volumetric fluorescence imaging.
  • Adding explicit physical constraints (for example, known dispersion relations) to the MLP input might permit even lower sampling fractions.
  • Running the adaptive sampler in a closed loop with the instrument could allow experiments to stop early once a target fidelity is reached.

Load-bearing premise

The multilayer perceptron accurately represents the mapping from coordinates to intensities without introducing systematic bias or erasing important oscillatory features.

What would settle it

A side-by-side comparison on the same sample showing that the 1/32-sample reconstruction deviates significantly from the fully sampled reference in peak positions, amplitudes, or oscillatory behavior.

Figures

Figures reproduced from arXiv: 2604.06561 by Chi-Jui Ho, Harsh Bhakta, Nicholas Antipa, Wei Xiong.

Figure 1
Figure 1. Figure 1: Hyper-spectrum reconstruction through neural rendering. We sparsely sample the hyper-spectrum collected from the interaction between ultrafast pulses and samples, and then use the proposed neural rendering framework to reconstruct the dense hyper-spectrum data in both oscillatory and non-oscillatory regimes. scanning necessitate hours to capture a single 2DIR data cube [6]; when extended across multiple wa… view at source ↗
Figure 2
Figure 2. Figure 2: Dimensionality of Spatially-Resolved 2DIR. a) The system diagram. b) Illustration of the four coordinate axes defining the hyperspectral volume. Each 2DIR frame parameterized by (𝜔1, 𝜔3), while the temporal and spatial variations are recorded in different waiting time 𝑡2 and spatial position 𝑥. 2. Materials and Methods 2.1. Theoretical Framework and Acquisition of Spatially-Resolved 2DIR HSI captures spect… view at source ↗
Figure 3
Figure 3. Figure 3: Our MLP-based neural representation model. We construct neural represen￾tations for signals subsampled along either the population (slow varying) or coherence (fast varying) time dimensions. Population: A coordinate-based MLP is trained to learn the temporal dynamics along population time from sparsely sampled data and to generalize this knowledge to unsampled points during inference. Coherence: The MLP is… view at source ↗
Figure 4
Figure 4. Figure 4: 2D representation of 4DIR data. A bounding box is defined around the dominant peak in each (𝜔1, 𝜔3) slice. The intensities enclosed within this region are averaged to reduce the four-dimensional data to a two-dimensional representation along the (𝑥, 𝑡2) axes, which is subsequently used for evaluation. Slowly Varying Intensities. Given that temporal dynamics along the population time axis (𝑡2) are inherentl… view at source ↗
Figure 5
Figure 5. Figure 5: Our loss-driven adaptive sampling method. Beginning with the initially sampled coordinates (blue dots), the MLP is trained to model the spectral dynamics and the loss is evaluated along the population time 𝑡2. The interval exhibiting the highest average prediction error is identified, and its midpoint is selected as the next sample. 2.4. Adaptive Sampling We apply a loss-aware adaptive sampling strategy [2… view at source ↗
Figure 6
Figure 6. Figure 6: Summary of results under different numbers of repeated samplings. (a) Measurements averaged over varying data averaging counts (𝑟) and reconstructed spectra under the same count. (c) Using the least noisy case (highest 𝑟) as reference, we compute the MSE of other measurements and predictions against it. (d) Mean and (e) standard deviation profiles of reconstructions at different repeat counts, with the ref… view at source ↗
Figure 7
Figure 7. Figure 7: Summary of results under different sampling intervals (SI) along the population time. (a) Reference and reconstructed spectra obtained at various sampling rates along the population time axis. (b–d) Peak intensity, mean and standard deviation profiles of reconstructions at different sampling intervals, with the reference shown as a solid black line [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Reconstruction results from subsampled 𝑡1 domain data. (a) Reconstructed 2DIR spectrum compared against the ground truth. (b) Temporal raw measurements showing the fully sampled ground truth and the subsampled input. (c) A 1D intensity profile extracted from (b) demonstrating signal recovery. To visualize oscillations beyond the native samples, the ground truth is densified via sinc interpolation, while th… view at source ↗
Figure 9
Figure 9. Figure 9: Reconstruction results from subsampled 𝑡1 domain data using different algorithms. (a) The 2DIR spectrum and (b) temporal measurements generated from different reconstruction algorithms compared against the ground truth [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Reconstruction performance at different 𝑡1 sampling rates. We display the reconstructed 2DIR spectrum under different sampling rates and associated temporal data. SR: sampling rate. Measurement Reconstruction Measurement Reconstruction [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Reconstruction 2DIR Spectrum under low accumulation and sampling rates. With 𝑟 = 5 and undersampled 𝑡1, 𝑡2 measurement, we display an unsampled 𝑡2 slice of 2DIR spectrum and a slice of (𝑡2, 𝜔3) projection. the oscillatory structures in the time domain are easily oversmoothed under high missing rates. Furthermore, while GIRAF can recover 2DIR data from low 𝑡1 sampling rates, it is unable to address undersa… view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of 2DIR Spectrum Reconstruction Under Differing Sampling Strategies. (a)-(c) Fully sampled temporal data and representative sampling masks for the two evaluated strategies. (d) Temporal reconstruction statistics across 10 independent trials. (e)-(g) Corresponding reconstructed 2DIR spectra for the examples shown in (a)-(c). (h) Summary of spectral reconstruction performance. (i)-(j) 1D illustra… view at source ↗
Figure 13
Figure 13. Figure 13: Reconstruction results from different sampling strategies. We overlay (a) peak-intensity, (b) mean, and (c) standard-deviation profiles derived from spectra reconstructed using different sampling strategies [PITH_FULL_IMAGE:figures/full_fig_p015_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Effect of statistical moment-matching. Spectral recovery results obtained with (orange) and without (blue) moment-matching constraints for (a) the mean and (b) the standard deviation profiles. 3.4. Statistical-Moment Constraint To validate the impact of the statistical-moment constraint detailed in Sec. 2.2.2, we performed an ablation study by excluding the statistical and physical constraints from the op… view at source ↗
read the original abstract

High-dimensional hyperspectral imaging (HSI) enables the visualization of ultrafast molecular dynamics and complex, heterogeneous spectra. However, applying this capability to resolve spatially varying vibrational couplings in two-dimensional infrared (2DIR) spectroscopy, a type of coherent multidimensional spectroscopy (CMDS), necessitates prohibitively long data acquisition, driven by dense Nyquist sampling requirements and the need for extensive signal accumulation. To address this challenge, we introduce a physics-informed neural representation approach that efficiently reconstructs dense spatially-resolved 2DIR hyperspectral images from sparse experimental measurements. In particular, we used a multilayer perceptron (MLP) to model the relationship between the sub-sampled 4D coordinates and their corresponding spectral intensities, and recover densely sampled 4D spectra from limited observations. The reconstruction results demonstrate that our method, using a fraction of the samples, faithfully recovers both oscillatory and non-oscillatory spectral dynamics in experimental measurement. Moreover, we develop a loss-aware adaptive sampling method to progressively introduce potentially informative samples for iterative data collection while conducting experiments. Experimental results show that the proposed approach achieves high-fidelity spectral recovery using only $1/32$ of the sampling budget, as opposed to exhaustive sampling, effectively reducing total experiment time by up to 32-fold. This framework offers a scalable solution for accelerating any experiments with hypercube data, including multidimensional spectroscopy and hyperspectral imaging, paving the way for rapid chemical imaging of transient biological and material systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a physics-informed neural representation using a multilayer perceptron (MLP) to model the mapping from sparsely sampled 4D coordinates (spatial, temporal, spectral) to spectral intensities in 2DIR hyperspectral imaging. Combined with a loss-aware adaptive sampling strategy for iterative data acquisition, the approach claims to reconstruct dense 4D spectra with high fidelity using only 1/32 of the full Nyquist sampling budget, yielding up to a 32-fold reduction in experimental acquisition time while preserving both oscillatory vibrational couplings and non-oscillatory dynamics.

Significance. If the reconstruction fidelity holds under rigorous validation, the method would offer a practical route to accelerate data collection in coherent multidimensional spectroscopy and other hypercube-based experiments, potentially enabling higher-throughput studies of transient molecular systems. The adaptive sampling component adds value by focusing acquisition on informative points, though its impact depends on the underlying representation quality.

major comments (2)
  1. [Abstract] Abstract: The central claim of 'high-fidelity spectral recovery' and 'faithfully recovers both oscillatory and non-oscillatory spectral dynamics' using 1/32 sampling is asserted without any quantitative metrics (RMSE, spectral correlation, error bars), baseline comparisons (e.g., to bilinear interpolation, compressed sensing, or standard NN reconstruction), or validation details on experimental 2DIR data; this absence directly limits assessment of the 32-fold speedup assertion.
  2. [Methods (neural representation)] MLP architecture description: The model is presented as a standard multilayer perceptron mapping 4D coordinates to intensities with no reference to positional encodings, Fourier feature mappings, or periodic activations. Given the well-known spectral bias of MLPs toward low-frequency components, this omission creates a concrete risk that high-frequency oscillatory features in vibrational 2DIR spectra will be attenuated or distorted, which is load-bearing for the fidelity claim.
minor comments (1)
  1. [Abstract] The abstract and introduction would benefit from a brief statement of the precise loss function used for the MLP and how the adaptive sampler selects points (e.g., uncertainty or gradient-based criteria).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We address each major comment point by point below, indicating the revisions made to strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of 'high-fidelity spectral recovery' and 'faithfully recovers both oscillatory and non-oscillatory spectral dynamics' using 1/32 sampling is asserted without any quantitative metrics (RMSE, spectral correlation, error bars), baseline comparisons (e.g., to bilinear interpolation, compressed sensing, or standard NN reconstruction), or validation details on experimental 2DIR data; this absence directly limits assessment of the 32-fold speedup assertion.

    Authors: We agree that the abstract would benefit from explicit quantitative support for the central claims. In the revised version, we have updated the abstract to include specific metrics (RMSE and spectral correlation coefficients with error bars) drawn from the experimental 2DIR reconstructions reported in the Results section. We have also added a concise reference to baseline comparisons against bilinear interpolation (with full details and additional comparisons to standard NN reconstruction provided in Section 4 and the supplementary material). These changes directly address the assessment of the 32-fold speedup while preserving the abstract's brevity. revision: yes

  2. Referee: [Methods (neural representation)] MLP architecture description: The model is presented as a standard multilayer perceptron mapping 4D coordinates to intensities with no reference to positional encodings, Fourier feature mappings, or periodic activations. Given the well-known spectral bias of MLPs toward low-frequency components, this omission creates a concrete risk that high-frequency oscillatory features in vibrational 2DIR spectra will be attenuated or distorted, which is load-bearing for the fidelity claim.

    Authors: We acknowledge the referee's valid concern about MLP spectral bias. Our physics-informed formulation incorporates a loss term that enforces consistency with the known structure of 2DIR spectra (including both oscillatory couplings and non-oscillatory dynamics), which empirically enables recovery of high-frequency features as demonstrated by the preserved vibrational cross-peaks in the experimental results. To strengthen the manuscript, we have added a dedicated paragraph in the Methods section discussing the spectral bias issue, the role of the physics-informed loss in mitigating it for this domain, and an ablation analysis confirming that high-frequency oscillatory content is retained at the reported sampling density. While we did not employ explicit Fourier feature mappings in the current implementation, the added discussion clarifies why the chosen architecture suffices for the fidelity achieved. revision: partial

Circularity Check

0 steps flagged

No significant circularity in neural representation or adaptive sampling

full rationale

The paper trains an MLP directly on measured sub-sampled 4D experimental data to reconstruct the dense hyperspectral volume and uses model loss to guide adaptive sampling of new measurements. This is a standard data-driven interpolation procedure with no self-definitional reduction, no fitted parameter renamed as an independent prediction, and no load-bearing self-citation chain or imported uniqueness theorem. The central claim rests on empirical fidelity to held-out experimental measurements rather than any algebraic identity or ansatz smuggled through prior work by the same authors.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The MLP and adaptive sampler likely contain architecture and loss hyperparameters that function as free parameters, but none are enumerated.

pith-pipeline@v0.9.0 · 5569 in / 1044 out tokens · 39567 ms · 2026-05-10T18:41:21.300766+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    we used a multilayer perceptron (MLP) to model the relationship between the sub-sampled 4D coordinates and their corresponding spectral intensities... four hidden layers, each comprising 64 neurons with Rectified Linear Unit (ReLU) activations... LMSE,slow ... Lmoment ... Lmono ... Lsmooth

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

53 extracted references · 4 canonical work pages · 2 internal anchors

  1. [1]

    Imaging spectrometry for earth remote sensing,

    A. F. Goetz, G. Vane, J. E. Solomon, and B. N. Rock, “Imaging spectrometry for earth remote sensing,” Science228, 1147–1153 (1985)

    1985

  2. [2]

    Hyperspectral imaging: techniques for spectral detection and classification,

    C.-I. Chang, “Hyperspectral imaging: techniques for spectral detection and classification,” Kluwer Acad. Publ. (2003)

    2003

  3. [3]

    Hyperspectral 2d ir imaging: Principles and applications to biological and materials systems,

    M. T. Zanni, “Hyperspectral 2d ir imaging: Principles and applications to biological and materials systems,” in2020 45th International Conference on Infrared, Millimeter, and Terahertz Waves (IRMMW-THz),(IEEE, 2020), pp. 1–2

    2020

  4. [4]

    Emerging trends in chemical applications of lasers,

    American Chemical Society, “Emerging trends in chemical applications of lasers,” (ACS Publications,2021)

    2021

  5. [5]

    Ultrafast 2d ir microscopy,

    C. R. Baiz, D. Schach, and A. Tokmakoff, “Ultrafast 2d ir microscopy,” Opt. Express22, 18724–18735 (2014)

    2014

  6. [6]

    Spatially resolved two-dimensional infrared spectroscopy via wide-field microscopy,

    J. S. Ostrander, A. L. Serrano, A. Ghosh, and M. T. Zanni, “Spatially resolved two-dimensional infrared spectroscopy via wide-field microscopy,” ACS Photonics3, 1315–1323 (2016)

    2016

  7. [7]

    Review of snapshot spectral imaging technologies,

    N. Hagen and M. W. Kudenov, “Review of snapshot spectral imaging technologies,” Opt. Eng.52, 090901–090901 (2013)

    2013

  8. [8]

    Single-shotcompressivespectralimagingwithadual-disperserarchitecture,

    M.E.Gehm,R.John,D.J.Brady,et al.,“Single-shotcompressivespectralimagingwithadual-disperserarchitecture,” Opt. express15, 14013–14027 (2007)

    2007

  9. [9]

    Compact implementation of fourier transform two-dimensional ir spectroscopy without phase ambiguity,

    J. Helbing and P. Hamm, “Compact implementation of fourier transform two-dimensional ir spectroscopy without phase ambiguity,” J. Opt. Soc. Am. B28, 171–178 (2010)

    2010

  10. [10]

    Accelerating two-dimensional infrared spectroscopy while preserving lineshapes using giraf,

    I. Bhattacharya, J. J. Humston, C. M. Cheatum, and M. Jacob, “Accelerating two-dimensional infrared spectroscopy while preserving lineshapes using giraf,” Opt. letters42, 4573–4576 (2017)

    2017

  11. [11]

    Compressedsensingformultidimensionalspectroscopyexperiments,

    J.N.Sanders,S.K.Saikin,S.Mostame,et al.,“Compressedsensingformultidimensionalspectroscopyexperiments,” The journal physical chemistry letters3, 2697–2702 (2012)

    2012

  12. [12]

    Accelerated 2d-ir using compressed sensing,

    J. A. Dunbar, D. G. Osborne, J. M. Anna, and K. J. Kubarych, “Accelerated 2d-ir using compressed sensing,” The J. Phys. Chem. Lett.4, 2489–2492 (2013)

    2013

  13. [13]

    Compressively sampled two-dimensional infrared spectroscopy that preserves line shape information,

    J. J. Humston, I. Bhattacharya, M. Jacob, and C. M. Cheatum, “Compressively sampled two-dimensional infrared spectroscopy that preserves line shape information,” The J. Phys. Chem. A121, 3088–3093 (2017)

    2017

  14. [14]

    Cutting through the noise: Extracting dynamics from ultrafast spectra using dynamic mode decomposition,

    C. Xu and C. R. Baiz, “Cutting through the noise: Extracting dynamics from ultrafast spectra using dynamic mode decomposition,” The J. Phys. Chem. A127, 9853–9862 (2023)

    2023

  15. [15]

    Generative adversarial neural networks for denoising coherent multidimensional spectra,

    Z. A. Al-Mualem and C. R. Baiz, “Generative adversarial neural networks for denoising coherent multidimensional spectra,” The J. Phys. Chem. A126, 3816–3825 (2022)

    2022

  16. [16]

    Learning implicit fields for generative shape modeling,

    Z. Chen and H. Zhang, “Learning implicit fields for generative shape modeling,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition,(2019), pp. 5939–5948

    2019

  17. [17]

    Deepsdf: Learning continuous signed distance functions for shape representation,

    J. J. Park, P. Florence, J. Straub,et al., “Deepsdf: Learning continuous signed distance functions for shape representation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition,(2019), pp. 165–174

    2019

  18. [18]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” J. Comput. physics378, 686–707 (2019)

    2019

  19. [19]

    Physics-informed machine learning: A survey on problems, methods and applications.arXiv preprint arXiv:2211.08064, 2022

    Z. Hao, S. Liu, Y. Zhang,et al., “Physics-informed machine learning: A survey on problems, methods and applications,” arXiv preprint arXiv:2211.08064 (2022)

    work page arXiv 2022

  20. [20]

    A differentiable wave optics model for end-to-end computational imaging system optimization,

    C.-J. Ho, Y. Belhe, S. Rotenberg,et al., “A differentiable wave optics model for end-to-end computational imaging system optimization,” inProceedings of the IEEE/CVF International Conference on Computer Vision,(2025), pp. 28042–28051

    2025

  21. [21]

    Acorn: Adaptive coordinate networks for neural representation,

    J. N. Martel, D. B. Lindell, C. Z. Lin,et al., “Acorn: Adaptive coordinate networks for neural representation,” ACM Trans. Graph. (SIGGRAPH) (2021)

    2021

  22. [22]

    Structure of the amide i band of peptides measured by femtosecond nonlinear-infrared spectroscopy,

    P. Hamm, M. Lim, and R. M. Hochstrasser, “Structure of the amide i band of peptides measured by femtosecond nonlinear-infrared spectroscopy,” The J. Phys. Chem. B102, 6123–6138 (1998)

    1998

  23. [23]

    Principles of nonlinear optical spectroscopy,

    S. Mukamel, “Principles of nonlinear optical spectroscopy,” ( Oxf. Univ. Press. (1995)

    1995

  24. [24]

    Enabling multiple intercavity polariton coherences by adding quantum confinement to cavity molecular polaritons,

    Z. Yang, H. H. Bhakta, and W. Xiong, “Enabling multiple intercavity polariton coherences by adding quantum confinement to cavity molecular polaritons,” Proc. National Acad. Sci.120, e2206062120 (2023)

    2023

  25. [25]

    Ultrafast coherence delocalization in real space simulated by polaritons

    Xiang, B., Yang, Z., You, Y. & Xiong, W. “Ultrafast coherence delocalization in real space simulated by polaritons.” Advanced Optical Materials.10, 2102237 (2022)

    2022

  26. [26]

    Two-dimensional infrared spectroscopy of vibrational polaritons

    Xiang, B., Ribeiro, R., Dunkelberger, A., Wang, J., Li, Y., Simpkins, B., Owrutsky, J., Yuen-Zhou, J. & Xiong, W. “Two-dimensional infrared spectroscopy of vibrational polaritons.”Proceedings Of The National Academy Of Sciences.115, 4845-4850 (2018)

    2018

  27. [27]

    Manipulating optical nonlinearities of molecular polaritons by delocalization

    Xiang, B., Ribeiro, R., Li, Y., Dunkelberger, A., Simpkins, B., Yuen-Zhou, J. & Xiong, W. “Manipulating optical nonlinearities of molecular polaritons by delocalization.”Science Advances.5, eaax5196 (2019)

    2019

  28. [28]

    Application of 2d ir bioimaging: Hyperspectral images of formalin-fixedpancreatictissuesandobservationofslowproteindegradation,

    S. S. Dicke, A. M. Alperstein, K. L. Schueler,et al., “Application of 2d ir bioimaging: Hyperspectral images of formalin-fixedpancreatictissuesandobservationofslowproteindegradation,”TheJ.Phys.Chem.B125,9517–9525 (2021)

    2021

  29. [29]

    Spatially-resolved fluorescence-detected two-dimensional electronic spectroscopy probes varying excitonic structure in photosynthetic bacteria,

    V. Tiwari, Y. A. Matutes, A. T. Gardiner,et al., “Spatially-resolved fluorescence-detected two-dimensional electronic spectroscopy probes varying excitonic structure in photosynthetic bacteria,” Nat. Commun.9, 4219 (2018)

    2018

  30. [30]

    Transient absorption microscopy: Technological innovations and applications in materials science and life science,

    Y. Zhu and J.-X. Cheng, “Transient absorption microscopy: Technological innovations and applications in materials science and life science,” The J. Chem. Phys.152(2020)

    2020

  31. [31]

    Progress and prospects in optical ultrafast microscopy in the visible spectral region: Transient absorption and two-dimensional microscopy,

    N. Gross, C. T. Kuhs, B. Ostovar,et al., “Progress and prospects in optical ultrafast microscopy in the visible spectral region: Transient absorption and two-dimensional microscopy,” The J. Phys. Chem. C127, 14557–14586 (2023)

    2023

  32. [32]

    & Sun, J

    He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition.Proceedings Of The IEEE Conference On Computer Vision And Pattern Recognition. pp. 770-778 (2016)

    2016

  33. [33]

    Deep Learning using Rectified Linear Units (ReLU)

    A. F. Agarap, “Deep learning using rectified linear units (relu),” arXiv preprint arXiv:1803.08375 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  34. [34]

    On the spectral bias of neural networks,

    N. Rahaman, A. Baratin, D. Arpit,et al., “On the spectral bias of neural networks,” inInternational conference on machine learning,(PMLR, 2019), pp. 5301–5310

    2019

  35. [35]

    Subdiffusive exciton transport in quantum dot solids

    Akselrod, G., Prins, F., Poulikakos, L., Lee, E., Weidman, M., Mork, A., Willard, A., Bulovic, V. & Tisdale, W. “Subdiffusive exciton transport in quantum dot solids.”Nano Letters.14, 3556-3562 (2014)

    2014

  36. [36]

    Resolving ultrafast exciton migration in organic solids at the nanoscale

    Penwell, S., Ginsberg, L., Noriega, R. & Ginsberg, N. “Resolving ultrafast exciton migration in organic solids at the nanoscale.”Nature Materials.16, 1136-1141 (2017)

    2017

  37. [37]

    Tuning the coherent propagation of organic exciton-polaritons through dark state delocalization

    Pandya, R., Ashoka, A., Georgiou, K., Sung, J., Jayaprakash, R., Renken, S., Gai, L., Shen, Z., Rao, A. & Musser, A. “Tuning the coherent propagation of organic exciton-polaritons through dark state delocalization.”Advanced Science. 9, 2105569 (2022)

    2022

  38. [38]

    From enhanced diffusion to ultrafast ballistic motion of hybrid light–matter excitations

    Balasubrahmaniyam, M., Simkhovich, A., Golombek, A., Sandik, G., Ankonina, G. & Schwartz, T. “From enhanced diffusion to ultrafast ballistic motion of hybrid light–matter excitations.”Nature Materials.22, 338-344 (2023)

    2023

  39. [39]

    Cavity-enhanced energy transport in molecular systems

    Sandik, G., Feist, J., García-Vidal, F. & Schwartz, T. “Cavity-enhanced energy transport in molecular systems. ” Nature Materials.24, 344-355 (2025)

    2025

  40. [40]

    Ultrafast imaging of polariton propagation and interactions

    Xu, D., Mandal, A., Baxter, J., Cheng, S., Lee, I., Su, H., Liu, S., Reichman, D. & Delor, M. “Ultrafast imaging of polariton propagation and interactions.”Nature Communications.14, 3881 (2023)

    2023

  41. [41]

    Solutions of ill-posed problems,

    R. A. Willoughby, “Solutions of ill-posed problems,” Siam Rev.21, 266 (1979)

    1979

  42. [42]

    R. C. Aster, B. Borchers, and C. H. Thurber,Parameter estimation and inverse problems(Elsevier, 2018)

    2018

  43. [43]

    Pytorch: An imperative style, high-performance deep learning library

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L. & Others “Pytorch: An imperative style, high-performance deep learning library.”Advances In Neural Information Processing Systems.32(2019)

    2019

  44. [44]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  45. [45]

    Fourier features let networks learn high frequency functions in low dimensional domains,

    M. Tancik, P. Srinivasan, B. Mildenhall,et al., “Fourier features let networks learn high frequency functions in low dimensional domains,” Adv. neural information processing systems33, 7537–7547 (2020)

    2020

  46. [46]

    Phase distortion-free paramagnetic nmr spectra,

    E. Ravera, “Phase distortion-free paramagnetic nmr spectra,” J. Magn. Reson. Open8, 100022 (2021)

    2021

  47. [47]

    A structured dictionary perspective on implicit neural representations,

    G. Yüce, G. Ortiz-Jiménez, B. Besbinar, and P. Frossard, “A structured dictionary perspective on implicit neural representations,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,(2022), pp. 19228–19238

    2022

  48. [48]

    Understanding and mitigating extrapolation failures in physics-informed neural networks,

    L. Fesser, L. D’Amico-Wong, and R. Qiu, “Understanding and mitigating extrapolation failures in physics-informed neural networks,” arXiv preprint arXiv:2306.09478 (2023)

    work page arXiv 2023

  49. [49]

    On the gibbs phenomenon and its resolution,

    D. Gottlieb and C.-W. Shu, “On the gibbs phenomenon and its resolution,” SIAM review39, 644–668 (1997)

    1997

  50. [50]

    Optimized reconstructions of compressively sampled two-dimensional infrared spectra,

    J. J. Humston, I. Bhattacharya, M. Jacob, and C. M. Cheatum, “Optimized reconstructions of compressively sampled two-dimensional infrared spectra,” The J. chemical physics150(2019)

    2019

  51. [51]

    Plenoctrees for real-time rendering of neural radiance fields,

    A. Yu, R. Li, M. Tancik,et al., “Plenoctrees for real-time rendering of neural radiance fields,” inProceedings of the IEEE/CVF international conference on computer vision,(2021), pp. 5752–5761

    2021

  52. [52]

    Learning continuous image representation with local implicit image function,

    Y. Chen, S. Liu, and X. Wang, “Learning continuous image representation with local implicit image function,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition,(2021), pp. 8628–8638

    2021

  53. [53]

    Nermo: Learningimplicitneuralrepresentationsfor3dhumanmotionprediction,

    D.Wei,H.Sun,X.Sun,andS.Hu,“Nermo: Learningimplicitneuralrepresentationsfor3dhumanmotionprediction,” inEuropean Conference on Computer Vision,(Springer, 2024), pp. 409–427

    2024