---
title: Feature Map Convergence Evaluation (FMCE)
url: https://www.emergentmind.com/topics/feature-map-convergence-evaluation-fmce
type: topic
---

# Feature Map Convergence Evaluation (FMCE)

Searching arXiv for papers directly related to Feature Map Convergence Evaluation and adjacent feature-map evaluation frameworks.
Feature Map Convergence Evaluation (FMCE) is an evaluation methodology for quantifying the convergence of feature maps in deep neural networks at the module level, with the stated aim of assessing how “well-trained” intermediate representations are and thereby enabling independent evaluation of functional modules rather than treating the model as an end-to-end black box [2405.04041]. In the literature represented here, FMCE appears in several closely related forms: as a feature-map analysis framework for autonomous driving perception models [2405.04041], as a closed-loop training framework that uses predicted convergence scores as auxiliary supervision [2508.06109], and as an extended quality-scoring formulation for autonomous driving models via feature map-truth similarity and dual-granularity weighting [2508.07552]. Related work also uses feature-map representations to assess training maturity or faithfulness through similarity, attribution, or dependency spectra, which places FMCE within a broader family of representation-centric evaluation methods [2409.11969], [2104.05527], [2305.20074].

## 1. Origins and problem setting

FMCE was introduced to address the absence of independent evaluation for functional modules in deep neural networks, particularly in perception systems whose internal mechanisms are otherwise difficult to inspect [2405.04041]. The motivating setting is that autonomous driving perception models are composed of multiple functional modules that interact through complex relationships, yet are predominantly optimized through end-to-end training. This creates difficulties for interpretability and optimization because intermediate modules are not evaluated independently [2405.04041].

The core proposal in "Feature Map Convergence Evaluation for Functional Module" [2405.04041] is a feature-map-based methodology comprising three components: the Convergence Quantification Indicator (CQI), the Feature Map Convergence Score (FMCS), and the Feature Map Convergence Evaluation Network (FMCE-Net). CQI provides a quantitative indicator of whether training has approached a flat-loss regime, FMCS assigns discrete convergence levels to extracted feature maps, and FMCE-Net predicts those convergence levels from feature maps alone [2405.04041].

Later work extends this initial diagnostic setting in two directions. "FMCE-Net++: Feature Map Convergence Evaluation and Training" [2508.06109] turns FMCE into a closed-loop training signal by attaching a pretrained, frozen FMCE-Net as an auxiliary head. "Decoupled Functional Evaluation of Autonomous Driving Models via Feature Map Quality Scoring" [2508.07552] generalizes the earlier feature-map-truth similarity approach into a Feature Map Quality Score (FMQS), combining model-level and feature-level evidence through a Dual-Granularity Dynamic Weighted Scoring System. This suggests that FMCE has evolved from a convergence diagnostic into a broader representation-quality supervision framework.

## 2. Formalization of convergence in the original FMCE framework

The original FMCE framework defines convergence through the behavior of the training loss curve [2405.04041]. The Convergence Quantification Indicator is constructed from a smoothed loss sequence, a first-order difference, and a moving average of absolute differences over a window of epochs:

$$
\widetilde{L}_m = \alpha L_{m-1} + (1 - \alpha) L_m
$$

$$
\Delta \widetilde{L}_m = \widetilde{L}_m - \widetilde{L}_{m-1}
$$

$$
\Delta \widetilde{L}_{m}^{avg} = \frac{1}{B} \sum_{b=1}^{B} |\Delta \widetilde{L}_{m-b+1}|
$$

The model is considered converged when

$$
CQI_m = \Delta \widetilde{L}_{m}^{avg} \leq \mu_{CQI}
$$

where $\alpha$ is typically set to $0.85$ [2405.04041]. CQI therefore operationalizes convergence as a recent-history average of smoothed loss variation rather than as a single-epoch criterion.

After identifying the training trajectory, FMCE divides convergence into $K$ convergence phases by applying a logarithmic transform to the smoothed loss and uniformly segmenting the total decrease in log-loss:

$$
\widetilde{L}_{m}^{log} = \log(\widetilde{L}_m)
$$

$$
G( \widetilde{L}^{log}) = \left| \widetilde{L}_{E_K}^{log} -\widetilde{L}_{E_0}^{log} \right|
$$

$$
\Delta G(\widetilde{L}^{log}) = \frac{G(\widetilde{L}^{log})}{K}
$$

If $C_{E_k}(x_n)$ denotes the feature map generated at epoch marker $E_k$ for sample $x_n$, then the Feature Map Convergence Score is defined as

$$
\text{FMCS} (C_{E_{k}(x_n)) = k
$$

with higher $k$ denoting higher convergence [2405.04041]. The result is a labeled dataset of feature maps indexed by convergence phase rather than by semantic class.

This formulation makes FMCS a discretized representation of training maturity. A plausible implication is that FMCE treats convergence as an intrinsic property of intermediate representations, not merely of end-task outputs.

## 3. FMCE-Net and supervised prediction of convergence levels

FMCE-Net is the predictive component of the original framework [2405.04041]. It is described as a lightweight CNN tailored to the size of the extracted feature maps, with $2$-$3$ convolutional layers, MaxPooling, ReLU, flattening, and fully connected layers that output a $K$-dimensional probability vector over convergence classes [2405.04041]. The training objective is standard multiclass cross-entropy over FMCS labels.

The later FMCE-Net++ paper gives an explicit probability-vector formulation. For each feature map, FMCE-Net predicts

$$
\hat{\mathbf{q}_i = (\hat{q}_{i1}, ..., \hat{q}_{iK})
$$

and is trained with

$$
\mathcal{L}_{\mathrm{CE}(\theta) = -\frac{1}{N}\sum_{i=1}^{N}\sum_{k=1}^{K} s_{ik}\,\log\!\bigl(\hat{q}_{ik}\bigr)
$$

where $s_{ik}=1$ if the $i$-th map belongs to class $k$, and $0$ otherwise [2508.06109]. The FMCS dataset is built by freezing a backbone at $K$ checkpoints, extracting feature maps for the full training set at each checkpoint, and using checkpoint index as the FMCS label [2508.06109].

The empirical validation reported for FMCE-Net in the original paper covers MNIST, CIFAR-10, and Mini-ImageNet with ResNet-50 and ShuffleNet v2 backbones [2405.04041]. Reported prediction results are summarized in the paper as follows:

| FMCS Dataset | Accuracy | Precision |
|---|---:|---:|
| MNIST-R | 0.9994 | 0.9994 |
| CIFAR-R | 0.9409 | 0.9418 |
| Imagenet-R | 0.9989 | 0.9989 |

The same table also reports an average accuracy of $0.9754$ with standard deviation $0.0354$ across the settings [2405.04041]. The paper further states that Grad-CAM visualizations become more concentrated and localized on semantically relevant object regions as FMCS increases, while lower FMCS corresponds to less focused attention [2405.04041]. This is presented as qualitative evidence that the score tracks the maturation of feature extraction.

## 4. Closed-loop training with FMCE-Net++

"FMCE-Net++: Feature Map Convergence Evaluation and Training" [2508.06109] extends FMCE from a post hoc evaluator to a training-time auxiliary supervisor. The framework attaches a pretrained and frozen FMCE-Net to a backbone so that intermediate feature maps are routed both to the primary classifier and to the FMCE auxiliary head [2508.06109].

The primary task loss is standard multiclass cross-entropy:

$$
\mathcal{L}_{\text{base} = -\frac{1}{N} \sum_{i=1}^N \sum_{c=1}^C y_{ic} \log \hat{p}_{ic}
$$

The FMCS loss uses a “fully converged” pseudo-label at class $K$:

$$
r_{ik} = \begin{cases} 1, & k = K \\ 0, & \text{otherwise} \end{cases}
$$

Source: https://www.emergentmind.com/topics/feature-map-convergence-evaluation-fmce