{
  "created": "2020-02-10T00:00:00+00:00",
  "revision_id": 75,
  "uuid": "f08101ea-fdc4-4bd6-862f-1f66a41c6ad7",
  "links": {
    "bibtex": "https://inspirehep.net/api/literature/1779199?format=bibtex",
    "latex-eu": "https://inspirehep.net/api/literature/1779199?format=latex-eu",
    "latex-us": "https://inspirehep.net/api/literature/1779199?format=latex-us",
    "json": "https://inspirehep.net/api/literature/1779199?format=json",
    "json-expanded": "https://inspirehep.net/api/literature/1779199?format=json-expanded",
    "cv": "https://inspirehep.net/api/literature/1779199?format=cv",
    "citations": "https://inspirehep.net/api/literature/?q=refersto%3Arecid%3A1779199"
  },
  "metadata": {
    "citation_count_without_self_citations": 24,
    "citation_count": 31,
    "core": false,
    "titles": [
      {
        "title": "Normalizing Flows on Tori and Spheres",
        "source": "arXiv"
      },
      {
        "title": "Normalizing Flows on Tori and Spheres",
        "source": "submitter"
      }
    ],
    "$schema": "https://inspirehep.net/schemas/records/hep.json",
    "authors": [
      {
        "uuid": "bac0011a-b2c3-4f82-9f41-d7c0964ea86b",
        "record": {
          "$ref": "https://inspirehep.net/api/authors/1873694"
        },
        "full_name": "Rezende, Danilo Jimenez",
        "signature_block": "RASANDd"
      },
      {
        "uuid": "e5b78773-b0ec-4572-9c5e-38d8bbed5397",
        "record": {
          "$ref": "https://inspirehep.net/api/authors/2563091"
        },
        "full_name": "Papamakarios, George",
        "signature_block": "PAPANACARg"
      },
      {
        "uuid": "c0fc1df8-03a9-4e03-80c5-11c2716951e6",
        "record": {
          "$ref": "https://inspirehep.net/api/authors/1873693"
        },
        "full_name": "Racanière, Sébastien",
        "signature_block": "RACANARs"
      },
      {
        "uuid": "c4848daf-b0b9-414e-be0c-e866de63b269",
        "record": {
          "$ref": "https://inspirehep.net/api/authors/1866886"
        },
        "full_name": "Albergo, Michael S.",
        "signature_block": "ALBARGm"
      },
      {
        "uuid": "bc4db2cb-564a-4a9e-996a-b7109f0f70b4",
        "record": {
          "$ref": "https://inspirehep.net/api/authors/1796339"
        },
        "full_name": "Kanwar, Gurtej",
        "signature_block": "CANWARg",
        "curated_relation": true
      },
      {
        "uuid": "8788e468-e7f3-4109-8817-76317302dc3e",
        "record": {
          "$ref": "https://inspirehep.net/api/authors/1091403"
        },
        "full_name": "Shanahan, Phiala E.",
        "signature_block": "SANANp",
        "curated_relation": true
      },
      {
        "uuid": "a26d3865-1af4-4ef2-ad38-c3134e6f6df5",
        "record": {
          "$ref": "https://inspirehep.net/api/authors/1012846"
        },
        "full_name": "Cranmer, Kyle",
        "signature_block": "CRANARk"
      }
    ],
    "curated": false,
    "figures": [
      {
        "key": "2752a3ee11bf1f3075c42f535db36771",
        "url": "https://inspirehep.net/files/2752a3ee11bf1f3075c42f535db36771",
        "source": "arxiv",
        "caption": "Illustration of the recursive flow on the sphere $\\sphere{2}$.",
        "filename": "flow_s2_diagram.png"
      },
      {
        "key": "87130ae5398169c64588ec9639a18c30",
        "url": "https://inspirehep.net/files/87130ae5398169c64588ec9639a18c30",
        "source": "arxiv",
        "caption": "Comparing flows on $\\mathbb{T}^2$ using ESS, for the targets in \\Cref{fig:flow_torus_densities} and various inverse temperatures $\\beta$. A larger value of $\\beta$ makes the target densities more concentrated and therefore harder to learn. Numbers in square brackets indicate the number of components $K$ used for each transform. All values are averages across $10$ runs with different neural-network initialization. See also \\Cref{fig:results_torus_all} (appendix) for KL values and further comparisons. {\\bf Top row}: Unimodal target. {\\bf Middle row}: Multi-modal target with $3$ mixture components. {\\bf Bottom row}: Correlated target.",
        "filename": "torus_ess.png"
      },
      {
        "key": "eef4adf2fcbcb38eb3dd49b25912baf2",
        "url": "https://inspirehep.net/files/eef4adf2fcbcb38eb3dd49b25912baf2",
        "source": "arxiv",
        "caption": "Learned densities on $\\torus{2}$ using NCP, M\\\"obius and CS flows. Densities shown on the torus are from NCP\\@.",
        "filename": "fixed_rot_all_high_res_3x5.png"
      },
      {
        "key": "32df937a05742e5b9b593b70e2f42f0d",
        "url": "https://inspirehep.net/files/32df937a05742e5b9b593b70e2f42f0d",
        "source": "arxiv",
        "caption": "Learned multi-modal density on $\\sphere{2}$ using exponential-map flows, using the Mollweide projection for visualization. The model is a composition of $24$ exponential-map transforms, using the radial scalar field with $1$ component.",
        "filename": "sphere_s2_4mode_density.png"
      },
      {
        "key": "9658d1cf81e566a71ed68a5a162fd6e9",
        "url": "https://inspirehep.net/files/9658d1cf81e566a71ed68a5a162fd6e9",
        "source": "arxiv",
        "caption": "Learned multi-modal density on $\\SU{2} \\cong \\sphere{3}$ using the recursive flow. Each column shows an $\\sphere{2}$ slice of the $\\sphere{3}$ density along a fixed axis using the Mollweide projection. \\textbf{Top row}: target density. \\textbf{Bottom row}: learned density. We used a M\\\"obius transform with $K_m=32$ for the circle, and spline transforms with $K_s = 64$ for the two intervals ($\\text{ESS}=84\\%$, $\\text{KL}=0.14$).",
        "filename": "su2_slices_smaller.png"
      },
      {
        "key": "d5602d3290bec93bab5363c72f7ab218",
        "url": "https://inspirehep.net/files/d5602d3290bec93bab5363c72f7ab218",
        "source": "arxiv",
        "caption": "Detailed illustration of the recursive flow on the sphere $\\sphere{D}$ showing the explicit wiring of the conditional flows. The sphere $\\sphere{D}$ is recursively transformed to the cylinder $\\sphere{1}\\times[-1, 1]^{D-1}$, then an autoregressive flow is applied to the cylinder, and finally the cylinder is transformed back to the sphere.",
        "filename": "flow_s2_detailed.png"
      },
      {
        "key": "f01028408b20a7edf630cac3c00b726b",
        "url": "https://inspirehep.net/files/f01028408b20a7edf630cac3c00b726b",
        "source": "arxiv",
        "caption": "Probability density functions of convex combinations of $15$ M\\\"obius transformations applied to a uniform base distribution on the circle $\\sphere{1}$. Each of these distributions required $30=15\\times 2$ parameters.",
        "filename": "expressive_moebius15.png"
      },
      {
        "key": "2fd81c6d9bced533869398d52e090dee",
        "url": "https://inspirehep.net/files/2fd81c6d9bced533869398d52e090dee",
        "source": "arxiv",
        "caption": "Same as \\Cref{fig.torus}, with KL and values for the Fourier transforms added. For the Fourier models, the numbers between brackets represent used frequencies, and a number before the bracket means each frequency was repeated. For example, Fourier$3[1-4]$ is a Fourier model with $12$ frequencies: $3$ frequencies of $k$ for each $k=1,\\ldots,4$.",
        "filename": "torus_all.png"
      },
      {
        "key": "dcef2ed8639853ac2c5745c94e72f866",
        "url": "https://inspirehep.net/files/dcef2ed8639853ac2c5745c94e72f866",
        "source": "arxiv",
        "caption": "Learning a non-axis-aligned density on $\\sphere{2}$ using \\Cref{eq:flow_sd_1,eq:flow_sd_2,eq:flow_sd_3} with and without composing with a learnable rotation. We compare M\\\"obius-spline flow (MS) ($K_s=32$, $K_m=12$), learnable rotation composed with MS (Rot $\\circ$ MS), small MS ($K_s=3$, $K_m=3$) (SMS) and learnable rotation composed with SMS (Rot $\\circ$ SMS)\\@. We observed no substantial differences between these models, suggesting that the particular choice of axis inside the flow has no impact on performance.",
        "filename": "slanted.png"
      },
      {
        "key": "03b75ee17ef12cd87029f2ef6be8e96e",
        "url": "https://inspirehep.net/files/03b75ee17ef12cd87029f2ef6be8e96e",
        "source": "arxiv",
        "caption": "Learning the posterior density over joint angles of a $6$-link $2$D robot arm. We used an autoregressive M\\\"obius flow on the torus $\\torus{6}$ to approximate the posterior density of joint angles resulting in a Gaussian mixture density for the tip of the robot arm. {\\bf Left}: The heat map shows the target density for the tip of the robot arm, a Gaussian mixture with two modes with centres at $(-0.5, 0.5)$ and $(0.6, -0.1)$. White paths show arm configurations sampled from the learned model in angle space converted to position space. {\\bf Right}: Density of the tips of the robot arm using samples from the learned model.",
        "filename": "robot_arm.png"
      },
      {
        "key": "f7be7f025d68f5655255c4039aefa8f4",
        "url": "https://inspirehep.net/files/f7be7f025d68f5655255c4039aefa8f4",
        "source": "arxiv",
        "caption": "Learning a complex density from data samples using autoregressive spline flows. {\\bf Left}: Target density from which i.i.d.\\ data samples were generated. {\\bf Middle}: Model samples overlaid on target density; {\\bf Right}: Heat map of the learned density.",
        "filename": "globe.png"
      }
    ],
    "license": [
      {
        "url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
        "license": "arXiv nonexclusive-distrib 1.0",
        "material": "preprint"
      }
    ],
    "texkeys": [
      "Rezende:2020hrd"
    ],
    "citeable": true,
    "abstracts": [
      {
        "value": "Normalizing flows are a powerful tool for building expressive distributions in high dimensions. So far, most of the literature has concentrated on learning flows on Euclidean spaces. Some problems however, such as those involving angles, are defined on spaces with more complex geometries, such as tori or spheres. In this paper, we propose and compare expressive and numerically stable flows on such spaces. Our flows are built recursively on the dimension of the space, starting from flows on circles, closed intervals or spheres.",
        "source": "submitter"
      },
      {
        "value": "Normalizing flows are a powerful tool for building expressive distributions in high dimensions. So far, most of the literature has concentrated on learning flows on Euclidean spaces. Some problems however, such as those involving angles, are defined on spaces with more complex geometries, such as tori or spheres. In this paper, we propose and compare expressive and numerically stable flows on such spaces. Our flows are built recursively on the dimension of the space, starting from flows on circles, closed intervals or spheres.",
        "source": "arXiv"
      }
    ],
    "references": [
      {
        "raw_refs": [
          {
            "value": "Arnaud, D., de Freitas, N., and Gordon, N. Sequential Monte Carlo methods in practice. Information Science and Statistics. Springer New York, 2001. Boomsma, W., Mardia, K. V., Taylor, C. C., FerkinghoffBorg, J., Krogh, A., and Hamelryck, T. A generative, probabilistic model of local protein structure. Proceedings of the National Academy of Sciences, 105(26):8932– 8937, 2008. Davidson, T. R., Falorsi, L., De Cao, N., Kipf, T., and Tomczak, J. M. Hyperspherical variational auto-encoders. Conference on Uncertainty in Artificial Intelligence, 2018. Dinh, L., Sohl-Dickstein, J., and Bengio, S. Density estimation using Real NVP. International Conference on Learning Representations, 2017. Downs, T. D. Orientation statistics. Biometrika, 59(3): 665–676, 1972. Downs, T. D. and Mardia, K. V. Circular regression. Biometrika, 89(3):683–698, 2002. Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G. Cubic-spline flows. ICML workshop on Invertible Neural Networks and Normalizing Flows, 2019a. Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G. Neural spline flows. Advances in Neural Information Processing Systems, 2019b.",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Sequential Monte Carlo methods in practice. Information Science and Statistics / New York, 2001. Boomsma, W., Mardia, K. V., Taylor, C. C., FerkinghoffBorg, J., Krogh, A., and Hamelryck, T. A / generative, probabilistic model of local protein structure / Davidson, T. R., Falorsi, L., De Cao, N., Kipf, T., and Tomczak, J. M / Hyperspherical variational auto-encoders. Conference on Uncertainty in Artificial Intelligence, 2018. Dinh, L., Sohl-Dickstein, J., and Bengio, S"
          ],
          "authors": [
            {
              "full_name": "Arnaud, D."
            },
            {
              "full_name": "de Freitas, N."
            },
            {
              "full_name": "Gordon, N."
            }
          ],
          "imprint": {
            "publisher": "Springer"
          },
          "publication_info": {
            "year": 2008,
            "page_end": "8937",
            "page_start": "8932",
            "journal_title": "Proc.Nat.Acad.Sci.",
            "journal_volume": "105"
          }
        }
      },
      {
        "raw_refs": [
          {
            "value": "Arnaud, D., de Freitas, N., and Gordon, N. Sequential Monte Carlo methods in practice. Information Science and Statistics. Springer New York, 2001. Boomsma, W., Mardia, K. V., Taylor, C. C., FerkinghoffBorg, J., Krogh, A., and Hamelryck, T. A generative, probabilistic model of local protein structure. Proceedings of the National Academy of Sciences, 105(26):8932– 8937, 2008. Davidson, T. R., Falorsi, L., De Cao, N., Kipf, T., and Tomczak, J. M. Hyperspherical variational auto-encoders. Conference on Uncertainty in Artificial Intelligence, 2018. Dinh, L., Sohl-Dickstein, J., and Bengio, S. Density estimation using Real NVP. International Conference on Learning Representations, 2017. Downs, T. D. Orientation statistics. Biometrika, 59(3): 665–676, 1972. Downs, T. D. and Mardia, K. V. Circular regression. Biometrika, 89(3):683–698, 2002. Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G. Cubic-spline flows. ICML workshop on Invertible Neural Networks and Normalizing Flows, 2019a. Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G. Neural spline flows. Advances in Neural Information Processing Systems, 2019b.",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Density estimation using Real NVP. International Conference on Learning Representations, 2017 / Orientation statistics"
          ],
          "authors": [
            {
              "full_name": "Downs, T.D."
            }
          ],
          "publication_info": {
            "year": 1972,
            "page_end": "676",
            "page_start": "665",
            "journal_title": "Biometrika",
            "journal_volume": "59"
          }
        }
      },
      {
        "raw_refs": [
          {
            "value": "Arnaud, D., de Freitas, N., and Gordon, N. Sequential Monte Carlo methods in practice. Information Science and Statistics. Springer New York, 2001. Boomsma, W., Mardia, K. V., Taylor, C. C., FerkinghoffBorg, J., Krogh, A., and Hamelryck, T. A generative, probabilistic model of local protein structure. Proceedings of the National Academy of Sciences, 105(26):8932– 8937, 2008. Davidson, T. R., Falorsi, L., De Cao, N., Kipf, T., and Tomczak, J. M. Hyperspherical variational auto-encoders. Conference on Uncertainty in Artificial Intelligence, 2018. Dinh, L., Sohl-Dickstein, J., and Bengio, S. Density estimation using Real NVP. International Conference on Learning Representations, 2017. Downs, T. D. Orientation statistics. Biometrika, 59(3): 665–676, 1972. Downs, T. D. and Mardia, K. V. Circular regression. Biometrika, 89(3):683–698, 2002. Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G. Cubic-spline flows. ICML workshop on Invertible Neural Networks and Normalizing Flows, 2019a. Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G. Neural spline flows. Advances in Neural Information Processing Systems, 2019b.",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Circular regression / Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G / Cubic-spline flows. ICML workshop on Invertible Neural Networks and Normalizing Flows, 2019a. Durkan, C., Bekasov, A., Murray, I., and Papamakarios, G / Neural spline flows. Advances in Neural Information Processing Systems, 2019b"
          ],
          "authors": [
            {
              "full_name": "Downs, T.D."
            },
            {
              "full_name": "Mardia, K.V."
            }
          ],
          "publication_info": {
            "year": 2002,
            "page_end": "698",
            "page_start": "683",
            "journal_title": "Biometrika",
            "journal_volume": "89"
          }
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres and a Cauchy family on the sphere. arXiv preprint arXiv:1510.07679, 2015. Kato, S., Shimizu, K., and Shieh, G. S. A circular–circular regression model. Statistica Sinica, 18:633–645, 2008. Kent, J. T. The Fisher–Bingham distribution on the sphere. Journal of the Royal Statistical Society. Series B (Methodological), 44(1):71–80, 1982. Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. International Conference for Learning Representations, 2015. Kingma, D. P., Salimans, T., Jozefowicz, R., Chen, X., Sutskever, I., and Welling, M. Improved variational inference with inverse autoregressive flow. Advances in Neural Information Processing Systems, 2016. Kobayashi, S. and Nomizu, K. Foundations of differential geometry, volume 1. Interscience Publishers, 1963. Liu, J. S. Monte Carlo strategies in scientific computing. Springer Science & Business Media, 2008. Mardia, K. V. and Jupp, P. E. Directional statistics. Wiley Series in Probability and Statistics. John Wiley & Sons, 2009. Mardia, K. V., Taylor, C. C., and Subramaniam, G. K. Protein bioinformatics and mixtures of bivariate von Mises",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Normalizing Flows on Tori and Spheres and a Cauchy family on the sphere. arXiv preprint / 2015 / circular-circular regression model / Kent, J. T / The Fisher-Bingham distribution on the sphere. Journal of the Royal Statistical Society. Series B / (Methodological), 44(1):71-80, 1982. Kingma, D. P / and Ba, J. Adam / : A method for stochastic optimization. International Conference for Learning Representations, 2015. Kingma, D. P., Salimans, T., Jozefowicz, R., Chen, X., Sutskever, I., and Welling, M / Improved variational inference with inverse autoregressive flow. Advances in Neural Information Processing Systems, 2016. Kobayashi, S. and Nomizu, K / Foundations of differential geometry, volume 1"
          ],
          "authors": [
            {
              "full_name": "Kato, S."
            },
            {
              "full_name": "Shimizu, K."
            },
            {
              "full_name": "Shieh, G.S.A"
            }
          ],
          "imprint": {
            "publisher": "Interscience"
          },
          "arxiv_eprint": "1510.07679",
          "publication_info": {
            "year": 2008,
            "page_end": "645",
            "page_start": "633",
            "journal_title": "Statist.Sinica",
            "journal_volume": "18"
          }
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres and a Cauchy family on the sphere. arXiv preprint arXiv:1510.07679, 2015. Kato, S., Shimizu, K., and Shieh, G. S. A circular–circular regression model. Statistica Sinica, 18:633–645, 2008. Kent, J. T. The Fisher–Bingham distribution on the sphere. Journal of the Royal Statistical Society. Series B (Methodological), 44(1):71–80, 1982. Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. International Conference for Learning Representations, 2015. Kingma, D. P., Salimans, T., Jozefowicz, R., Chen, X., Sutskever, I., and Welling, M. Improved variational inference with inverse autoregressive flow. Advances in Neural Information Processing Systems, 2016. Kobayashi, S. and Nomizu, K. Foundations of differential geometry, volume 1. Interscience Publishers, 1963. Liu, J. S. Monte Carlo strategies in scientific computing. Springer Science & Business Media, 2008. Mardia, K. V. and Jupp, P. E. Directional statistics. Wiley Series in Probability and Statistics. John Wiley & Sons, 2009. Mardia, K. V., Taylor, C. C., and Subramaniam, G. K. Protein bioinformatics and mixtures of bivariate von Mises",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Publishers / Monte Carlo strategies in scientific computing"
          ],
          "authors": [
            {
              "full_name": "Liu, J.S."
            }
          ],
          "imprint": {
            "publisher": "Springer"
          },
          "publication_info": {
            "year": 1963
          }
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres and a Cauchy family on the sphere. arXiv preprint arXiv:1510.07679, 2015. Kato, S., Shimizu, K., and Shieh, G. S. A circular–circular regression model. Statistica Sinica, 18:633–645, 2008. Kent, J. T. The Fisher–Bingham distribution on the sphere. Journal of the Royal Statistical Society. Series B (Methodological), 44(1):71–80, 1982. Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. International Conference for Learning Representations, 2015. Kingma, D. P., Salimans, T., Jozefowicz, R., Chen, X., Sutskever, I., and Welling, M. Improved variational inference with inverse autoregressive flow. Advances in Neural Information Processing Systems, 2016. Kobayashi, S. and Nomizu, K. Foundations of differential geometry, volume 1. Interscience Publishers, 1963. Liu, J. S. Monte Carlo strategies in scientific computing. Springer Science & Business Media, 2008. Mardia, K. V. and Jupp, P. E. Directional statistics. Wiley Series in Probability and Statistics. John Wiley & Sons, 2009. Mardia, K. V., Taylor, C. C., and Subramaniam, G. K. Protein bioinformatics and mixtures of bivariate von Mises",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Science & Business Media / Directional statistics"
          ],
          "authors": [
            {
              "full_name": "Mardia, K.V."
            },
            {
              "full_name": "Jupp, P.E."
            }
          ],
          "imprint": {
            "publisher": "Wiley"
          },
          "publication_info": {
            "year": 2008
          }
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres and a Cauchy family on the sphere. arXiv preprint arXiv:1510.07679, 2015. Kato, S., Shimizu, K., and Shieh, G. S. A circular–circular regression model. Statistica Sinica, 18:633–645, 2008. Kent, J. T. The Fisher–Bingham distribution on the sphere. Journal of the Royal Statistical Society. Series B (Methodological), 44(1):71–80, 1982. Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. International Conference for Learning Representations, 2015. Kingma, D. P., Salimans, T., Jozefowicz, R., Chen, X., Sutskever, I., and Welling, M. Improved variational inference with inverse autoregressive flow. Advances in Neural Information Processing Systems, 2016. Kobayashi, S. and Nomizu, K. Foundations of differential geometry, volume 1. Interscience Publishers, 1963. Liu, J. S. Monte Carlo strategies in scientific computing. Springer Science & Business Media, 2008. Mardia, K. V. and Jupp, P. E. Directional statistics. Wiley Series in Probability and Statistics. John Wiley & Sons, 2009. Mardia, K. V., Taylor, C. C., and Subramaniam, G. K. Protein bioinformatics and mixtures of bivariate von Mises",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Science & Business Media / Series in Probability and Statistics. John / & Sons,. Mardia, K. V., Taylor, C. C., and Subramaniam, G. K / Protein bioinformatics and mixtures of bivariate von Mises"
          ],
          "authors": [
            {
              "full_name": "Mardia, K.V."
            },
            {
              "full_name": "Jupp, P.E."
            }
          ],
          "imprint": {
            "publisher": "Wiley"
          },
          "publication_info": {
            "year": 2009
          }
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres A. Density Transformations on Manifolds In this section, we explain how to update the density of a distribution transformed from one Riemannian manifold to another by a smooth map. We only consider the case where both manifolds are sub-manifolds of Euclidean spaces. Let M and N be D-dimensional manifolds embedded into Euclidean spaces Rm and Rn respectively. For example, M and N could be SD embedded in RD+1 as in Section 2.3. Both manifolds inherit a Riemannian metric from their embedding spaces. Let T be a smooth injective map T : M → N. We will assume that T can be extended to a smooth map between open neighbourhoods of the embedding spaces that contain M and N, and that we have chosen such an extension. For example, the exponential-map flow in Equation (19) can be written using the coordinates of the embedding space RD+1 , and can thus be extended to open neighbourhoods of the embedding spaces as desired. In what follows, we will use the fact that if u1, . . . , uD are",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Normalizing Flows on Tori and Spheres A. Density Transformations on Manifolds In this section, we explain how to update the density of a distribution transformed from one Riemannian manifold to another by a smooth map. We only consider the case where both manifolds are sub-manifolds of Euclidean spaces / and N be D-dimensional manifolds embedded into Euclidean spaces Rm and Rn respectively. For example, M and N could be SD embedded in RD+1 as in Section 2.3. Both manifolds inherit a Riemannian metric from their embedding spaces. Let T / be a smooth injective map T : M → N. We / will assume that T can be extended to a smooth map between open neighbourhoods of the embedding spaces that contain M and N, and that we have chosen such an extension. For example, the exponential-map flow in Equation (19) can be written using the coordinates of the embedding space RD+1, and can thus be extended to open neighbourhoods of the embedding spaces as desired. In what follows, we will use the fact that if u1,..., uD are"
          ],
          "authors": [
            {
              "full_name": "Let"
            }
          ]
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres At first, Proposition 1 might seem worrying since the density ratio in that proposition vanishes when r is −1 or 1. So, as r approaches the boundary of the interval [−1, 1], it seems that the correction term to the density will tend to infinity and lead to numerical instability. What saves us is that we do not use Tc→s on its own, and instead combine it with a particular flow transformation on SD−1 × [−1, 1] and the inverse Ts→c, as shown in Equations (12) to (14). In these formulas, the map g is a spline on the interval [−1, 1] which maps −1 to −1, 1 to 1, and has strictly positive slopes g0 (−1) and g0 (1). Looking only at −1 (the case 1 can be similarly dealt with), this means g(−1 + ) ≈ −1 + g0 (−1). As  goes to 0, the density corrections coming from Tc→s and Ts→c combine to  D 2 − 1",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Normalizing Flows on Tori and Spheres At first, Proposition 1 might seem worrying since the density ratio in that proposition vanishes when r is -1 or 1. So, as r approaches the boundary of the interval [-1, 1], it seems that the correction term to the density will tend to infinity and lead to numerical instability. What saves us is that we do not use Tc→s on its own, and instead combine it with a particular flow transformation on SD-1 × [-1, 1] and the inverse Ts→c, as shown in Equations (12) to (14). In these formulas, the map g is a spline on the interval [-1, 1] which maps -1 to -1, 1 to 1, and has strictly positive slopes g0 (-1) and g0 (1). Looking only at -1 (the case 1 can be similarly dealt with), this means g(-1 + ) ≈ -1 + g0 (-1). As goes to 0, the density corrections coming from Tc→s and Ts→c combine to D 2 - 1"
          ]
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres Figure 7. Probability density functions of convex combinations of 15 M¨ obius transformations applied to a uniform base distribution on the circle S1 . Each of these distributions required 30 = 15 × 2 parameters. Target Expression Parameters Unimodal pA(θ1, θ2) ∝ exp[cos(θ1 − φ1) + cos(θ2 − φ2)] φ = (4.18, 5.96) Multi-modal pB(θ1, θ2) ∝ 1 3 P3 i=1 pA(θ1, θ2; φi) φ = {(0.21, 2.85), (1.89, 6.18), (3.77, 1.56)} Correlated pC(θ1, θ2) ∝ exp[cos(θ1 + θ2 − φ)] φ = 1.94 Table 2. Target densities used for experiments on T2 . Figure 8. Same as Figure 2, with KL and values for the Fourier transforms added. For the Fourier models, the numbers between brackets represent used frequencies, and a number before the bracket means each frequency was repeated. For example, Fourier3[1 − 4] is a Fourier model with 12 frequencies: 3 frequencies of k for each k = 1, . . . , 4.",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Normalizing Flows on Tori and Spheres Figure 7. Probability density functions of convex combinations of 15 Möbius transformations applied to a uniform base distribution on the circle S1. Each of these distributions required 30 = 15 × 2 parameters. Target Expression Parameters Unimodal pA(θ1, θ2) ∝ exp[cos(θ1 - φ1) + cos(θ2 - φ2)] φ = (4.18, 5.96) Multi-modal pB(θ1, θ2) ∝ 1 3 P3 i=1 pA(θ1, θ2"
          ]
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres Figure 7. Probability density functions of convex combinations of 15 M¨ obius transformations applied to a uniform base distribution on the circle S1 . Each of these distributions required 30 = 15 × 2 parameters. Target Expression Parameters Unimodal pA(θ1, θ2) ∝ exp[cos(θ1 − φ1) + cos(θ2 − φ2)] φ = (4.18, 5.96) Multi-modal pB(θ1, θ2) ∝ 1 3 P3 i=1 pA(θ1, θ2; φi) φ = {(0.21, 2.85), (1.89, 6.18), (3.77, 1.56)} Correlated pC(θ1, θ2) ∝ exp[cos(θ1 + θ2 − φ)] φ = 1.94 Table 2. Target densities used for experiments on T2 . Figure 8. Same as Figure 2, with KL and values for the Fourier transforms added. For the Fourier models, the numbers between brackets represent used frequencies, and a number before the bracket means each frequency was repeated. For example, Fourier3[1 − 4] is a Fourier model with 12 frequencies: 3 frequencies of k for each k = 1, . . . , 4.",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "φi) φ = {(0.21, 2.85), (1.89, 6.18), (3.77, 1.56)} Correlated pC(θ1, θ2) ∝ exp[cos(θ1 + θ2 - φ)] φ = 1.94 Table 2. Target densities used for experiments on T2. Figure 8. Same as Figure 2, with KL and values for the Fourier transforms added. For the Fourier models, the numbers between brackets represent used frequencies, and a number before the bracket means each frequency was repeated. For example, Fourier3[1 - 4] is a Fourier model with 12 frequencies: 3 frequencies of k for each k = 1,..., 4"
          ]
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres coordinates, and x ∈ R3 is a point on the embedded sphere in Euclidean coordinates. On SU(2) ∼ = S3 , the target was a mixture of the same form where µ1 = (1.7, −1.5, 2.3), µ2 = (−3.0, 1.0, 3.0), µ3 = (0.6, −2.6, 4.5), µ4 = (−2.5, 3.0, 5.0), and x ∈ R4 is a point on the embedded sphere in Euclidean coordinates. G. Misaligned Density on S2 The recursive formulas shown in Equations (12) to (14) require choosing a sequence of axes in order to construct the cylindrical coordinate system. This may introduce artifacts to the density related to this choice of axes. To test if this results in numerical problems, we compare the flow from Equations (12) to (14) on a target density that forms a nonaxis-aligned ring against a composition of the same flow with a learned rotation. The results of this experiment are shown in Figure 9. We compared both large (Ks = 32, Km = 12) and small (Ks = 3, Km = 3) versions of the auto-regressive M¨ obiusSpline flow and observed no significant differences between",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Normalizing Flows on Tori and Spheres coordinates, and x ∈ R3 is a point on the embedded sphere in Euclidean coordinates. On SU(2) ∼ = S3, the target was a mixture of the same form where µ1 = (1.7, -1.5, 2.3), µ2 = (-3.0, 1.0, 3.0), µ3 = (0.6, -2.6, 4.5), µ4 = (-2.5, 3.0, 5.0), and x ∈ R4 is a point on the embedded sphere in Euclidean coordinates / on S2 The recursive formulas shown in Equations (12) to (14) require choosing a sequence of axes in order to construct the cylindrical coordinate system. This may introduce artifacts to the density related to this choice of axes. To test if this results in numerical problems, we compare the flow from Equations (12) to (14) on a target density that forms a nonaxis-aligned ring against a composition of the same flow with a learned rotation. The results of this experiment are shown in Figure 9. We compared both large (Ks = 32, Km = 12) and small (Ks = 3, Km = 3) versions of the auto-regressive MöbiusSpline flow and observed no significant differences between"
          ],
          "authors": [
            {
              "full_name": "Density"
            },
            {
              "full_name": "Misaligned, G."
            }
          ]
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres configurations of this arm are points in T6 . The position rk of a joint k = 1, . . . , 6 of the robot arm is given by rk = rk−1 +  lk cos   X j≤k θj  , lk sin   X j≤k θj    , where r0 = (0, 0) is the position where the arm is affixed,",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Normalizing Flows on Tori and Spheres configurations of this arm are points in T6. The position rk of a joint k = 1,..., 6 of the robot arm is given by rk = rk-1 +  lk cos   X j≤k θj  , lk sin   X j≤k θj    , where r0 = (0, 0) is the position where the arm is affixed"
          ]
        }
      },
      {
        "raw_refs": [
          {
            "value": "Normalizing Flows on Tori and Spheres",
            "schema": "text"
          }
        ],
        "reference": {
          "misc": [
            "Normalizing Flows on Tori and Spheres"
          ]
        }
      }
    ],
    "public_notes": [
      {
        "value": "Accepted to the International Conference on Machine Learning (ICML)\n  2020",
        "source": "arXiv"
      }
    ],
    "arxiv_eprints": [
      {
        "value": "2002.02428",
        "categories": [
          "stat.ML",
          "cs.LG"
        ]
      }
    ],
    "document_type": [
      "article"
    ],
    "preprint_date": "2020-02-06",
    "control_number": 1779199,
    "legacy_version": "20200703083622.0",
    "number_of_pages": 15,
    "inspire_categories": [
      {
        "term": "Other"
      },
      {
        "term": "Computing"
      },
      {
        "term": "Data Analysis and Statistics"
      }
    ],
    "legacy_creation_date": "2020-02-10"
  },
  "updated": "2023-03-06T19:51:29.799054+00:00",
  "id": "1779199"
}