# Landmark Distance-Profile Adapter

- ID: 1568
- Canonical URL: https://synthcore.org/idea/1568/landmark-distance-profile-adapter
- API JSON: https://synthcore.org/api/idea/1568.json
- API Markdown: https://synthcore.org/api/idea/1568.md
- Verification status: unverified
- Source: [arXiv:2607.28981](https://arxiv.org/abs/2607.28981)
- Category: architecture
- Solves: accuracy, sample-efficiency, scalability
- ML areas: embedding, graph-nn, conditioning
- Math tags: metric-geometry, functional-analysis, linear-algebra
- Ratings: usefulness 6/10; difficulty 5/10; novelty 6/10

## Idea description

Add a metric-aware front end that represents an arbitrary object x by its distances to a fixed set of reference objects rather than forcing x into a Euclidean or Hilbert embedding. Feed the resulting profile through a learned projection and concatenate it with the ordinary neural representation. This should be useful for graphs, trees, distributions, and sets where generic vectorization loses geometry or requires an expensive object-specific encoder.

## Mathematical statement

Definition 1 defines the distance profile embedding as \(\Phi_X(X):\Omega\to L_2(\Omega_X,\lambda_X)\), with \(\Phi_X(x)(u)=d_X(u,x)\). Here \((\Omega_X,d_X)\) is the object metric space, \(u\in\Omega_X\) is a reference object, \(\lambda_X\) is a measure over reference objects, and \(L_2(\Omega_X,\lambda_X)\) is the space of square-integrable real functions. The paper claims that this map is injective and preserves full distributional information under mild conditions, without requiring an isometric Hilbert embedding of \(\Omega_X\). Approximate the function by an empirical landmark measure \(\hat\lambda_K=\sum_{j=1}^K w_j\delta_{a_j}\), where \(a_j\) are reference objects, \(w_j\geq 0\), and \(\sum_j w_j=1\). The finite DPE is \(\hat\Phi_K(x)=[\sqrt{w_1}d_X(a_1,x),\ldots,\sqrt{w_K}d_X(a_K,x)]\), whose Euclidean inner product approximates the L2 inner product \(\langle\Phi_X(x),\Phi_X(y)\rangle_{L_2}=\int_{\Omega_X}d_X(u,x)d_X(u,y)\,d\lambda_X(u)\). The transferable property is geometric and distributional identifiability through distance profiles; the finite approximation is improved by increasing K or selecting more representative landmarks.

## Key formulas

- $$\Phi_X(X):\Omega\to L_2(\Omega_X,\lambda_X),\qquad \omega\mapsto\Phi_X(X(\omega))$$
- $$\Phi_X(x)(u)=d_X(u,x)$$
- $$\hat\lambda_K=\sum_{j=1}^{K}w_j\delta_{a_j},\qquad \hat\Phi_K(x)=\left[\sqrt{w_1}d_X(a_1,x),\ldots,\sqrt{w_K}d_X(a_K,x)\right]$$
- $$\langle\Phi_X(x),\Phi_X(y)\rangle_{L_2}=\int_{\Omega_X}d_X(u,x)d_X(u,y)\,d\lambda_X(u)\approx\sum_{j=1}^{K}w_jd_X(a_j,x)d_X(a_j,y)$$

## Implementation notes

Integrate the adapter immediately after object loading and before the main neural encoder. Inputs are objects x from a metric space with a callable distance function d_X(x,y); examples include graph edit distance, Wasserstein distance between empirical distributions, tree distance, or a point-cloud metric. First construct a landmark bank A={a_1,...,a_K} from the training set. Use uniform weights w_j=1/K initially, then optionally use stratified weights that approximate the training distribution. For each minibatch object x_i, compute D[i,j]=d_X(x_i,a_j), form P[i,j]=sqrt(w_j)*D[i,j], and standardize each landmark coordinate using training-set mean and variance. Pass P through a two-layer projection MLP and concatenate the projected profile with any existing object features.

Pseudocode:
landmarks = select_diverse_training_objects(K)
mu, sigma = estimate_coordinate_stats(landmarks, training_subset)
for x in minibatch:
    p = [sqrt(w[j]) * distance(x, landmarks[j]) for j in range(K)]
    p = (p - mu) / (sigma + eps)
    h = projection_mlp(p)
    prediction = network(h, other_inputs)
loss = task_loss(prediction, target)
backpropagate(loss)

The paper supplies the infinite-dimensional map and its injective, distribution-preserving rationale; the finite landmark approximation, coordinate normalization, landmark selection, and projection are engineering choices. Start with K=64 or 128 and compare random landmarks against farthest-point sampling under d_X. The first cheap experiment should use a small graph-classification benchmark with graph edit distance or a shortest-path-based graph metric, comparing the DPE adapter against a bag-of-statistics MLP and a standard graph encoder at matched parameter count. Success means higher test accuracy or calibration at equal training FLOPs, better performance when training data are reduced, and graceful scaling as K increases. Also test whether reducing K from 128 to 32 causes less degradation than reducing the dimension of a learned baseline embedding.

## Disclaimer

AI-generated research hypothesis, automatically tested. Not peer-reviewed.