setup chapter rework wip]

thesis/Main.tex

@@ -1138,36 +1138,20 @@ As can be seen, the Efficient encoder requires an order of magnitude fewer param
 {isoforest, ocsvm adapted, for ocsvm only dim reduced feasible (ae from deepsad)}
 {compared methods known $\rightarrow$ what methods were used}
 
-To contextualize the performance of DeepSAD, we compare against two widely used baselines: Isolation Forest and One-Class SVM (OCSVM). Both are included in the original DeepSAD codebase and the associated paper, and they represent well-understood but conceptually different families of anomaly detection. In our setting, the raw input dimensionality ($2048 \times 32$ per frame) is too high for a direct OCSVM fit, so we reuse the DeepSAD autoencoder’s \emph{encoder} as a learned dimensionality reduction (to the same latent size as DeepSAD). This choice is motivated by practicality (compute) and inductive bias: the encoder captures non-linear, domain-specific structure of the LiDAR range images, which linear methods like PCA may miss. \todo[inline]{Optionally add a sentence comparing PCA vs. AE and/or a cite.}
+
+To contextualize the performance of DeepSAD, we compare against two widely used baselines: Isolation Forest and One-Class SVM (OCSVM). Both are included in the original DeepSAD codebase and the associated paper, and they represent well-understood but conceptually different families of anomaly detection. In our setting, the raw input dimensionality ($2048 \times 32$ per frame) is too high for a direct OCSVM fit, so we reuse the DeepSAD autoencoder’s \emph{encoder} as a learned dimensionality reduction (to the same latent size as DeepSAD). This choice is motivated by practicality (compute) and inductive bias: the encoder captures non-linear, domain-specific structure of the LiDAR range images, which linear methods like PCA may miss. Together, these two baselines cover complementary perspectives: tree-based partitioning (Isolation Forest) and kernel-based boundary learning (OCSVM), providing a broad and well-established basis for comparison.
 
 Isolation Forest is an ensemble method for anomaly detection that builds on the principle that anomalies are easier to separate from the rest of the data. It constructs many binary decision trees, each by recursively splitting the data at randomly chosen features and thresholds. In this process, the “training” step consists of building the forest of trees: each tree captures different random partitions of the input space, and together they form a diverse set of perspectives on how easily individual samples can be isolated.
 
-Once trained, the method assigns an anomaly score to new samples by measuring their average path length through the trees. Normal samples, being surrounded by other similar samples, typically require many recursive splits and thus end up deep in the trees. Anomalies, by contrast, stand out in one or more features, which means they can be separated much earlier and end up closer to the root. The shorter the average path length, the more anomalous the sample is considered.
+Once trained, the method assigns an anomaly score to new samples by measuring their average path length through the trees. Normal samples, being surrounded by other similar samples, typically require many recursive splits and thus end up deep in the trees. Anomalies, by contrast, stand out in one or more features, which means they can be separated much earlier and end up closer to the root. The shorter the average path length, the more anomalous the sample is considered. This makes Isolation Forest highly scalable and robust: training is efficient since no explicit density estimation is required, and the resulting model is fast to apply to new data. In our setup, we apply Isolation Forest directly to the LiDAR input representation, providing a strong non-neural baseline for comparison against DeepSAD.
 
-This approach makes Isolation Forest highly scalable and robust: training is efficient since no explicit density estimation is required, and the resulting model is fast to apply to new data. In our setup, we apply sklearn's IsolationForest directly to the LiDAR input representation, providing a strong non-neural baseline for comparison against DeepSAD. We configured the method with its default parameters, namely 100 evaluators (trees) and 'auto' for samples, which in our case will randomly select 256 samples per evaluator.
-
-%Isolation Forest detects anomalies by \emph{isolation} rather than by modeling density. It builds an ensemble of randomly split trees; points that are easier to isolate—requiring fewer splits on average—are deemed more anomalous. Intuitively, rare or “different” samples tend to fall into shallow leaves because a few random thresholds already separate them from the bulk of the data. Practically, IF scales well to large datasets, is robust to high-dimensional noise when combined with a reasonable feature space, and requires minimal hyperparameter tuning (number of trees, subsample size). In our experiments, IF operates on the same feature representation as DeepSAD’s latent space to ensure direct comparability. \todo[inline]{Add implementation details such as number of estimators, subsampling size, random seed.}
-
-OCSVM is a popular method for anomaly detection that tackles the problem in a different way than tree-based methods like Isolation Forest. Instead of explicitly trying to separate normal and anomalous points during training, OCSVM is trained only on data assumed to be normal. The goal is to learn a boundary that encloses the majority of these training samples in such a way that new points lying outside this boundary can be identified as anomalies.
+While Isolation Forest relies on random partitioning of the input space, OCSVM takes a very different approach by learning a flexible boundary around normal samples. OCSVM is trained only on data assumed to be normal, with the goal of enclosing the majority of these samples in such a way that new points lying outside this boundary can be identified as anomalies.
 
 The boundary itself is learned using the support vector machine framework. In essence, OCSVM looks for a hyperplane in some feature space that maximizes the separation between the bulk of the data and the origin. To make this possible even when the normal data has a complex, curved shape, OCSVM uses a kernel function such as the radial basis function (RBF). The kernel implicitly maps the input data into a higher-dimensional space, where the cluster of normal samples becomes easier to separate with a simple hyperplane. When this separation is mapped back to the original input space, it corresponds to a flexible, nonlinear boundary that can adapt to the structure of the data.
 
 During training, the algorithm balances two competing objectives: capturing as many of the normal samples as possible inside the boundary, while keeping the region compact enough to exclude potential outliers. Once this boundary is established, applying OCSVM is straightforward — any new data point is checked against the learned boundary, with points inside considered normal and those outside flagged as anomalous.
 
-%One-Class SVM fits a decision boundary that encloses the majority of the data—treated as “normal”—and labels points outside as anomalies. Using a kernel (typically RBF), OCSVM learns a tight support around the data manifold; the $\nu$ parameter controls the expected fraction of outliers and margin errors. OCSVM can be very effective when an appropriate feature space is used, but it becomes computationally demanding and memory-intensive on high-dimensional raw inputs. Therefore, we apply OCSVM \emph{after} reducing dimensionality with the pretrained encoder to the same latent dimension used by DeepSAD. This yields a tractable problem size and focuses OCSVM on semantically meaningful, non-linear features rather than raw pixels. \todo[inline]{Record kernel, $\nu$, $\gamma$ selection strategy; e.g., grid search or heuristics.}
-
-\paragraph{Implementation notes and evaluation protocol.}
-We adapted the baseline implementations to our data loader and input format \todo[inline]{briefly describe file layout / preprocessing}, and added support for multiple evaluation targets per frame (two labels per data point), reporting both results per experiment. For OCSVM, the dimensionality reduction step is \emph{always} performed with the corresponding DeepSAD encoder and its autoencoder pretraining weights that match the evaluated setting (i.e., same latent size and backbone). Isolation Forest is trained on the same latent features for fairness. This protocol keeps feature spaces aligned across methods and isolates the effect of the anomaly detector itself. \todo[inline]{Add cross-validation / split details and metrics (e.g., ROC-AUC, PR-AUC) used for comparison.}
-
-\newsubsubsectionNoTOC{Training procedure (k‑fold cross‑validation, semi‑supervised loss) and hyperparameter choices}
-\newsubsubsectionNoTOC{Evaluation metrics (ROC, PRC, AUC, F1) and inference protocol}
-
-\threadtodo
-{what evaluation methods were used}
-{we know what is compared but want to know exactly how}
-{explain roc, prc, inference with experiment left out of training}
-{experiment overview given $\rightarrow$ details to deepsad during training?}
-
+We adapted the baseline implementations to our data loader and input format \todo[inline]{briefly describe file layout / preprocessing}, and added support for multiple evaluation targets per frame (two labels per data point), reporting both results per experiment. For OCSVM, the dimensionality reduction step is \emph{always} performed with the corresponding DeepSAD encoder and its autoencoder pretraining weights that match the evaluated setting (i.e., same latent size and backbone). Both baselines, like DeepSAD, output continuous anomaly scores. This allows us to evaluate them directly without committing to a fixed threshold.
 
 \section{Experiment Matrix \& Computational Environment}
 % Combines: Experiment Matrix + Hardware & Runtimes