---
title: "Invertible Transport Between Sample-Only... | SI Paper Desk"
url: https://thesimind.si/papers/2611.26187/
description: "PCF learns a reversible map between a dataset and a prior known only through samples. It matches kernel MMD in accuracy, uses less memory, and needs no..."
---

Number 12 of 12 · November 22, 2026

# Invertible Transport Between Sample-Only Distributions via Projected Characteristic Functions

PCF learns a reversible map between a dataset and a prior known only through samples. It matches kernel MMD in accuracy, uses less memory, and needs no density, adversarial training or paired data.

Juan Gutiérrez, Victor Gutiérrez-García and José Luis Blanco-Murillo · Submitted November 19, 2026 · arXiv 2611.26187v1 · cs.LG

[Read the paper on arXiv](https://arxiv.org/abs/2611.26187) · [The PDF](https://arxiv.org/pdf/2611.26187) · [The day's 12](https://thesimind.si/papers/day/2026-11-22/)

## What they found

The authors propose Projected Characteristic-Function matching (PCF). It learns an invertible map between a data distribution and a prior that can only be sampled, with no density formula. It compares the two distributions using projected empirical characteristic functions, which are summaries of a sample's shape. This avoids density estimation, adversarial training and paired observations. The authors also add an optional regularizer that picks transformations that move points less. PCF matches kernel MMD on the controlled benchmark and on raw-pixel MNIST-USPS, at a cost linear in the batch size. It reached lower endpoint error than MMD in 13 of 15 paired runs across library sizes. It cut discrepancy memory by 29.8% and trained at batch sizes where kernel MMD exceeds GPU memory.

## Why it matters

Many useful priors, such as real-world collections or simulator outputs, can be sampled but have no usable density. Standard normalizing flows (models that reverse a map between distributions) need that density. A method that works from samples alone widens what these models can use. The abstract reports gains mainly when data or memory is limited.

## How it changes the race

This is an incremental but practical step. It could lower the memory cost of aligning distributions, which helps teams with smaller GPUs. The abstract does not claim gains in frontier model capability.

### The abstract, by the authors

Many useful latent distributions can be sampled but do not have a tractable density. Such distributions may be represented by empirical collections of real-world observations or generated by scientific simulators and black-box generative models. These sample-only priors cannot be used directly in likelihood-based normalizing flows, which require evaluating the prior density. We propose Projected Characteristic-Function matching (PCF), a density-free framework for learning an invertible map between a data distribution and a prior specified only through samples. Our method aligns distributions using projected empirical characteristic functions, eliminating the need for density estimation, adversarial training, or paired observations. To address the non-uniqueness of the transport map, we introduce an optional transport regularizer that selects lower-displacement transformations while preserving endpoint alignment. We benchmark PCF on controlled transports, sample-efficiency studies, and high-dimensional alignment tasks. PCF matches the accuracy of kernel MMD on the controlled benchmark and on raw-pixel MNIST-USPS at a cost linear in the batch size, and it is more sample-efficient, achieving lower endpoint error than MMD in 13 of 15 paired runs across library sizes. By avoiding pairwise kernel evaluation, it reduces discrepancy memory by 29.8% and trains at batch sizes where kernel MMD exceeds GPU memory. These results show that PCF provides accurate and efficient invertible transport, particularly in sample- and memory-limited settings.

Quoted from arXiv:2611.26187. The research and its words belong to its authors.

Desk score **53** out of 100

Why it was picked

- Alignment
- Compute
- Evals
- Cross-listed in 2 fields

Topics

- [Efficiency](https://thesimind.si/papers/?tag=efficiency)
- [Theory](https://thesimind.si/papers/?tag=theory)
- [Training](https://thesimind.si/papers/?tag=training)
- [Data](https://thesimind.si/papers/?tag=data)
- [Vision](https://thesimind.si/papers/?tag=vision)
- [Benchmarks](https://thesimind.si/papers/?tag=benchmarks)

**arXiv ID:** 2611.26187v1
**Submitted:** November 19, 2026
**Fields:** cs.LG (Machine Learning), cs.CV (Computer Vision)
**Level:** Advanced

Fingerprint, SHA-256

`dad5af0240b244631b589e6b119ac8c553f9f7600f37aa82bacab03462b64c84`

In SI MIND's journal since Nov 22, 2026, 02:01 UTC. It goes on Solana with the next hourly Proof of Mind seal.

### The exact text fingerprinted

{"arxiv":"2611.26187","version":1,"title":"Invertible Transport Between Sample-Only Distributions via Projected Characteristic Functions","authors":["Juan Gutiérrez","Victor Gutiérrez-García","José Luis Blanco-Murillo"],"day":"2026-11-22","rank":12,"score":53,"status":"written","summary":{"takeaway":"PCF learns a reversible map between a dataset and a prior known only through samples. It matches kernel MMD in accuracy, uses less memory, and needs no density, adversarial training or paired data.","found":"The authors propose Projected Characteristic-Function matching (PCF). It learns an invertible map between a data distribution and a prior that can only be sampled, with no density formula. It compares the two distributions using projected empirical characteristic functions, which are summaries of a sample's shape. This avoids density estimation, adversarial training and paired observations. The authors also add an optional regularizer that picks transformations that move points less. PCF matches kernel MMD on the controlled benchmark and on raw-pixel MNIST-USPS, at a cost linear in the batch size. It reached lower endpoint error than MMD in 13 of 15 paired runs across library sizes. It cut discrepancy memory by 29.8% and trained at batch sizes where kernel MMD exceeds GPU memory.","matters":"Many useful priors, such as real-world collections or simulator outputs, can be sampled but have no usable density. Standard normalizing flows (models that reverse a map between distributions) need that density. A method that works from samples alone widens what these models can use. The abstract reports gains mainly when data or memory is limited.","race":"This is an incremental but practical step. It could lower the memory cost of aligning distributions, which helps teams with smaller GPUs. The abstract does not claim gains in frontier model capability.","safety":"","level":"advanced","tags":["efficiency","theory","training","data","vision","benchmarks"]}}

[Check it on Verify a thought](https://thesimind.si/proof-of-mind/)

## The authors

Juan Gutiérrez, Victor Gutiérrez-García, José Luis Blanco-Murillo

Credit for this research belongs to its authors. SI MIND's page is a plain-English reading of their abstract on arXiv, not a review of the full paper.

Related on the desk

## More papers on these topics

[Nov 17, 2026 · Number 12ParetoM³: Learning on the Pareto Set under Preference Guidance via Min-Max-Min OptimizationParetoM³ is a method for training SI models on several conflicting goals at once, steering toward the trade-off you prefer while staying on the set of best possible trade-offs.TrainingTheoryEfficiency](https://thesimind.si/papers/2611.18252/) · [Nov 18, 2026 · Number 2Nemotron 3.5 Bolt E4B 4GB: An Open and Efficient Small Language Model for Fast Conversational AgentsNVIDIA's authors post-trained a small open model from Gemma-4-E4B-it for fast local assistants. They report the top average accuracy on 21 benchmarks and faster decoding with a drafter model.EfficiencyLanguage modelsAgents](https://thesimind.si/papers/2611.20771/) · [Nov 15, 2026 · Number 4daVinci-Prover: Agent-Native Mid-Training for Formal Theorem ProvingThe authors build a 33.7B-token training corpus and a Lean-based agent harness for theorem proving, and report a 13.7 point average gain over a matched Qwen3-8B baseline.MathReasoningAgents](https://thesimind.si/papers/2611.16433/)

[Number 11](https://thesimind.si/papers/2611.25845/) · [All 12 for Nov 22, 2026](https://thesimind.si/papers/day/2026-11-22/)
