Fix mismatched citations flagged in citecheck audit

content/03.overview.md

@@ -117,7 +117,7 @@ When the parameters being partitioned define a network, the support itself can j
 
 ### Fine-tuned Models and Transfer Learning
 
-Another practical strategy for handling heterogeneity is fine-tuning. A global population model is first estimated, and then a smaller set of parameters is updated for particular subpopulations. This idea underlies transfer learning, where large pre-trained models are adapted to new tasks with limited additional training [@doi:10.48550/arXiv.2206.02058]. Fine-tuning balances the bias–variance tradeoff by borrowing statistical strength from large datasets while preserving flexibility for local adaptation. This notion was already recognized in early VCM literature as a form of semi-parametric estimation [@doi:10.1214/aos/1017939139].
+Another practical strategy for handling heterogeneity is fine-tuning. A global population model is first estimated, and then a smaller set of parameters is updated for particular subpopulations. This idea underlies transfer learning, where large pre-trained models are adapted to new tasks with limited additional training [@doi:10.1109/TKDE.2009.191]. Fine-tuning balances the bias–variance tradeoff by borrowing statistical strength from large datasets while preserving flexibility for local adaptation. This notion was already recognized in early VCM literature as a form of semi-parametric estimation [@doi:10.1214/aos/1017939139].
 
 ### Models for Explicit Subgroup Separation
 
@@ -134,4 +134,4 @@ Context-aware models can be organized along a spectrum of assumptions about the
 
 ![A spectrum of context awareness in modeling, showing global, grouped, smooth, and latent models](images/spectrum_context.png){#fig:spectrum-context width="70%"}
 
-Each formulation encodes different beliefs about parameter variation. The next section formalizes these principles and examines general strategies for adaptivity in statistical modeling. For a discussion of how subpopulation shifts influence generalization, see [@arXiv:2106.04486].
+Each formulation encodes different beliefs about parameter variation. The next section formalizes these principles and examines general strategies for adaptivity in statistical modeling. 

content/05.principles.md

@@ -25,7 +25,7 @@ A canonical example is the mixture-of-experts framework, where a gating network
 ### 4. Adaptivity implies selectivity
 Adaptation must not occur indiscriminately. Overreacting to noise leads to overfitting, defeating the purpose of adaptation. Selectivity provides the discipline that ensures adaptive mechanisms respond only when supported by reliable evidence.
 
-Classical statistics formalized this principle through methods such as Lepski’s rule for bandwidth selection, which balances bias and variance in nonparametric estimation [@doi:10.1214/aos/1030741083]. Aggregation methods such as the weighted majority algorithm show how selective weighting of multiple models can improve robustness [@doi:10.1006/inco.1994.1009]. In modern machine learning, Bayesian rules can activate test-time updates only when uncertainty is manageable [@doi:10.48550/arXiv.2410.03306], while confidence-based strategies prevent unstable adjustments by holding back adaptation under weak signals [@doi:10.48550/arXiv.2310.04941]. Sparse expert models apply the same principle architecturally, activating only a few experts for easy inputs but engaging more capacity for difficult cases [@doi:10.48550/arXiv.2403.07652]. These safeguards demonstrate that good adaptation is selective adaptation.
+Classical statistics formalized this principle through methods such as Lepski’s rule for bandwidth selection, which balances bias and variance in nonparametric estimation [@doi:10.1214/aos/1030741083]. Aggregation methods such as the weighted majority algorithm show how selective weighting of multiple models can improve robustness [@doi:10.1006/inco.1994.1009]. In modern machine learning, Bayesian rules can activate test-time updates only when uncertainty is manageable [@doi:10.48550/arXiv.2410.03306], while confidence-based strategies prevent unstable adjustments by holding back adaptation under weak signals [@doi:10.48550/arXiv.2204.02610]. Sparse expert models apply the same principle architecturally, activating only a few experts for easy inputs but engaging more capacity for difficult cases [@doi:10.48550/arXiv.2403.07652]. These safeguards demonstrate that good adaptation is selective adaptation.
 
 ### 5. Adaptivity is bounded by data efficiency
 Even with flexibility, heterogeneity, modularity, and selectivity in place, the scope of adaptation is fundamentally constrained by the availability of data. Fine-grained adaptation requires sufficient samples to estimate context-specific effects reliably. When data are scarce, adaptive systems risk inflating variance, capturing noise, or overfitting to idiosyncratic patterns. This limitation transcends individual methods and reflects a general statistical truth.
@@ -74,7 +74,7 @@ The efficiency of context-adaptive methods hinges on several key design principl
 
 Context-aware efficiency often relies on sparsity assumptions that limit the number of context-dependent parameters. This can be achieved through group sparsity, which encourages entire groups of context-dependent parameters to be zero simultaneously [@doi:10.1111/j.1467-9868.2005.00532.x], hierarchical regularization that applies different regularization strengths to different levels of context specificity [@doi:10.1111/j.2517-6161.1996.tb02080.x; @doi:10.1017/CBO9780511790942], and adaptive thresholding that dynamically adjusts sparsity levels based on context complexity.
 
-Efficient context-adaptive inference can be achieved through computational strategies that allocate resources based on context. Early stopping terminates optimization early for contexts where convergence is rapid [@doi:10.48550/arXiv.1606.04838], while context-dependent sampling uses different sampling strategies for different contexts [@doi:10.48550/arXiv.1809.09582]. Caching and warm-starting leverage solutions from similar contexts to accelerate optimization, particularly effective when contexts exhibit smooth variation [@doi:10.1561/2200000016].
+Efficient context-adaptive inference can be achieved through computational strategies that allocate resources based on context. Early stopping terminates optimization early for contexts where convergence is rapid [@doi:10.48550/arXiv.1606.04838]. Context-dependent sampling uses different sampling strategies for different contexts [@doi:10.18653/v1/2024.findings-naacl.122]. Caching and warm-starting leverage solutions from similar contexts to accelerate optimization, particularly effective when contexts exhibit smooth variation [@doi:10.1561/2200000016].
 
 The design of context-aware methods often involves balancing computational efficiency with interpretability. Linear context functions are more interpretable but may require more parameters, while explicit context encoding improves interpretability but may increase computational cost. Local context modeling provides better interpretability but may be less efficient for large-scale applications. These trade-offs must be carefully considered based on the specific requirements of the application domain, as demonstrated in recent work on adaptive optimization methods [@doi:10.48550/arXiv.1412.6980].
 

content/10.invariant.md

@@ -51,9 +51,3 @@ A simple and highly effective heuristic is Just Train Twice (JTT) [@arXiv:2107.0
 Providing a more formalized framework, Environment Inference for Invariant Learning (EIIL) aims to bridge the gap between unlabeled data and invariant learning algorithms like IRM [@arXiv:2010.07249]. Similar to JTT, EIIL begins by training a standard ERM model. It then uses the biases of this reference model to automatically partition the dataset into several inferred "environments." For instance, examples the model confidently gets right might form one environment, while those it gets wrong form another. These algorithmically generated environment labels can then be fed into any off-the-shelf invariant learning method to train a final, robust model. EIIL essentially transforms the problem from one requiring manual labels to one where environments can be discovered directly from the data itself.
 Collectively, these approaches demonstrate a continuum from fully supervised environment-aware optimization to self-supervised environment discovery, unified under the goal of achieving context-invariant generalization. Together, these methods illustrate a clear progression from fully-supervised techniques to more practical approaches that cleverly infer hidden data structure, all aiming to build models that are more robust and invariant to challenging shifts in context.
 
-
-
-<!-- - Just Train Twice: Improving Group Robustness without Training Group Information [@arXiv:2002.10384]
-- Environment Inference for Invariant Learning [@arXiv:2110.14048]
-- Distributionally Robust Neural Networks for Group Shifts [@arXiv:1911.08731] -->
-

content/11.applications_tools.md

@@ -36,9 +36,9 @@ Such contextualized networks have been applied in TCGA cancer genomics to identi
 
 Evaluating context-adaptive models requires careful consideration of predictive accuracy, robustness to variability, and scalability, with the emphasis varying by domain. Key aspects of performance evaluation include the choice of metrics, the handling of uncertainty, and assessment under stress or rare-event conditions.
 
-In healthcare, evaluation prioritizes patient-specific predictive accuracy and calibrated uncertainty. Common metrics include mean squared error (MSE), concordance indices (C-index), and calibration curves, which measure how well models capture longitudinal patient trajectories and provide reliable uncertainty estimates. Multi-target Bayesian approaches and survival models demonstrate the importance of capturing correlations across outcomes and assessing credible interval coverage to quantify predictive confidence [@arXiv:2509.08183; @arXiv:2509.01794]. Evaluations in this domain also highlight trade-offs between model complexity, interpretability, and computational feasibility, since high-fidelity patient-level predictions can be costly to compute.
+In healthcare, evaluation prioritizes patient-specific predictive accuracy and calibrated uncertainty. Common metrics include mean squared error (MSE), concordance indices (C-index), and calibration curves, which measure how well models capture longitudinal patient trajectories and provide reliable uncertainty estimates. Multi-target Bayesian approaches and survival models demonstrate the importance of capturing correlations across outcomes and assessing credible interval coverage to quantify predictive confidence [@arXiv:2509.01794]. Evaluations in this domain also highlight trade-offs between model complexity, interpretability, and computational feasibility, since high-fidelity patient-level predictions can be costly to compute.
 
-In finance and macro forecasting, performance evaluation emphasizes predictive accuracy under volatile conditions and resilience to structural breaks. Metrics such as root mean squared forecast error (RMSFE), log-likelihood, and stress-test performance are commonly used to assess how well models handle crises or abrupt shifts in data [@arXiv:2409.08354; @arXiv:2508.10055]. Probabilistic metrics, including posterior predictive checks and uncertainty bounds, provide additional insight into the reliability of forecasts, while chaos-informed diagnostics can highlight vulnerabilities to extreme events [@arXiv:2406.12274].
+In finance and macro forecasting, performance evaluation emphasizes predictive accuracy under volatile conditions and resilience to structural breaks. Metrics such as root mean squared forecast error (RMSFE), log-likelihood, and stress-test performance are commonly used to assess how well models handle crises or abrupt shifts in data [@arXiv:2409.08354; @arXiv:2508.10055]. Probabilistic metrics, including posterior predictive checks and uncertainty bounds, provide additional insight into the reliability of forecasts, while chaos-informed diagnostics can highlight vulnerabilities to extreme events [@arXiv:2509.08183].
 
 Across domains, consistent patterns emerge. Context-adaptive models outperform static baselines when variability is structured and partially predictable, but require sufficient examples of the desired adaptation behavior across contexts. Even if the dataset is large, if there are insufficient contexts to show context-specific adaptation, the model will fail to generalize across contexts. Thus, learning context-adaptive models requires special attention not only to the number of samples, but also to the number of distinct contexts and the distribution of contexts that samples are drawn from [@doi:10.1073/pnas.2411930122].
 

content/12.future_trends.md

@@ -21,7 +21,7 @@ While LoRA is traditionally learned via task-specific optimization, recent work
 Together, these approaches illustrate how adaptation can be achieved both flexibly and efficiently, balancing generalization and computational constraints.
 
 #### Bridging with Statistical and Causal Reasoning
-An emerging research direction integrates the representational capacity of foundation models with the rigor of statistical and causal inference. Language models can already extract relational patterns from text to propose or critique causal graphs [@doi:10.48550/arXiv.2305.07171]. Methods such as LMPriors treat foundation models as task-specific priors, improving sample efficiency in estimation and decision making [@doi:10.48550/arXiv.2210.12530]. Models can also generate natural-language rationales that clarify predictions and summarize statistical findings, enhancing interpretability and transparency [@doi:10.48550/arXiv.2310.05797].
+An emerging research direction integrates the representational capacity of foundation models with the rigor of statistical and causal inference. Methods such as LMPriors treat foundation models as task-specific priors, improving sample efficiency in estimation and decision making [@doi:10.48550/arXiv.2210.12530]. Models can also generate natural-language rationales that clarify predictions and summarize statistical findings, enhancing interpretability and transparency [@doi:10.48550/arXiv.2310.05797].
 
 Consequently, foundation models serve as bridges between flexible representation learning and principled inference, offering a path toward adaptive systems that are both data-efficient and theoretically grounded.