Pith. sign in

REVIEW 3 major objections 6 minor 202 references

ST-LoRA: Single Trajectory LoRA Ensemble for Uncertainty Aware Agricultural Segmentation

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read ST-LoRA shows a single fine-tuning run can yield a full uncertainty ensemble by swapping low-rank adapters over a frozen backbone.

desk verdict Efficiency and FF-layer findings are solid, but the 'matches full-rank ensembles' claim is tested only against a snapshot baseline and Table 4's parameter counts are mislabeled—two fixes before the abstract can be trusted. read the letter →

arxiv 2608.01530 v1 pith:5NABIJ2N submitted 2026-08-02 cs.CV

Mohamed Farag , Genc Hoxha , Yahia Maleki , Chris McCool , Ribana Roscher This is my paper
classification cs.CV
keywords LoRAsnapshotensemblinguncertaintyquantificationsemanticsegmentationdigitalagriculturecalibrationdistributionshiftout-of-distributiondetection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that a single fine-tuning run is enough to build a full uncertainty-aware ensemble for semantic segmentation: instead of training M independent models, freeze one pretrained backbone and fine-tune only low-rank adapters, saving one adapter set per cosine-annealing cycle and averaging their predictions at inference. It claims this single-trajectory LoRA ensemble (ST-LoRA) matches or beats a full-rank snapshot ensemble in mean IoU and calibration (ECE) on two agricultural datasets—cauliflower in open field and sweet pepper in glasshouse—using SegFormer and Mask2Former, while cutting trainable parameters, checkpoint size, and loading latency by one to two orders of magnitude. A second, revisionist claim is that in vision transformers used for dense prediction the feed-forward MLP layers, not the attention projections, are the LoRA modules that carry most of the task adaptation, contrary to the attention-only convention inherited from language models. If correct, the method gives agricultural robotics practical access to calibrated pixel-wise uncertainty at edge-deployment cost.

What carries the argument

The machinery is the LoRA reparameterization $\Delta \theta = (\alpha/r) B A$ injected into frozen transformer layers, combined with snapshot ensembling via a cosine-annealing learning-rate schedule. The effective adapter magnitude $\delta_r = \alpha/r$ acts as a single knob governing how strongly the adapter output blends with the frozen weights; the paper shows that small $\delta_r$ preserves the pretrained prior while large $\delta_r$ distorts it, and that this knob is a training-free calibration lever at deployment. The load-bearing property is the size asymmetry between small adapters and full checkpoints, which converts parameter savings directly into I/O and memory savings during ense

What would settle it

Re-run the BUP20 and GrowliFlower-L experiments with M=4 or M=5 independently initialized, fully fine-tuned Deep Ensemble members (Lakshminarayanan et al., 2017) using the paper's own hyperparameters. If that deep ensemble beats ST-LoRA by more than the paper's reported ST-LoRA-vs-FRE margin in ECE or mIoU, the claim that ST-LoRA matches full-rank ensembles fails. Additionally, an audit of Table 4's trainable-parameter counts against the base model (SegFormer-B2/B4, Mask2Former-Swin) would settle the 'under 10%' efficiency claim.

Watch

Extended reading notes

Core claim

ST-LoRA constructs M ensemble members from a single training trajectory by keeping a pretrained backbone frozen and inserting LoRA adapters into selected encoder layers (and, in some configurations, decoder query/key projections). A cosine-annealing schedule with warm restarts visits distinct minima, and one adapter checkpoint is saved at the end of each cycle; the predictive posterior is the uniform mixture of the M adapted models (Eq. 10). The paper's central empirical claim is that this low-rank snapshot ensemble matches or exceeds the full-rank snapshot ensemble (FRE, defined as a fully fine-tuned model snapshotted the same way) in both mIoU and ECE on GrowliFlower-L and BUP20, while red

Load-bearing premise

The central claim that a single-trajectory LoRA ensemble matches full-rank quality is established only against a snapshot ensemble of one fully fine-tuned model ('FRE'), not against the standard Deep Ensemble of independently initialized full models; if the latter baseline is the intended meaning of 'full-rank ensemble,' the paper's central trade-off claim is unsupported.

Editorial extensions

If this is right

  • Calibrated uncertainty maps for crop segmentation become deployable on mid-range edge devices: five 5.5 MB adapters replace four 430 MB checkpoints, cutting startup overhead from seconds to milliseconds on a Raspberry Pi 4B.
  • Ensemble diversity can be obtained from a single optimization run, reducing the training cost of an M-member ensemble by about a factor of M compared with M independent full model runs.
  • The $\delta_r = \alpha/r$ scaling ratio offers a deployment-time calibration knob: practitioners can tune adapter strength after fine-tuning without retraining.
  • For dense prediction with vision transformers, LoRA modules should be placed in feed-forward (MLP) layers first; attention-only adaptation is both less accurate and worse calibrated in this regime.
  • Low-rank ensemble members show lower cross-seed variance in calibration and OoD detection than full-rank baselines, suggesting more stable uncertainty estimates for field deployment.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural extension the authors do not run: compare ST-LoRA against a true Deep Ensemble of independently initialized full models on the same seeds. My expectation is that the accuracy gap will be small but nonzero, and the calibration gap will depend on whether the shared-backbone constraint suppresses diversity exactly where it matters.
  • The feed-forward-layer finding appears transferable: since MLP layers act as key-value memories across modalities, the same LoRA placement rule likely holds for depth estimation, panoptic segmentation, and other dense transformers; a quick check on Cityscapes or ADE20K would confirm.
  • Because all members share one backbone, epistemic uncertainty measured by adapter disagreement is confined to a low-rank subspace; full-rank sources of model uncertainty (e.g., head collapse, decoder misalignment) stay invisible. A testable extension is to measure how much of the deep-ensemble gap is recovered by increasing rank r toward full rank.
  • The authors' own numbers suggest 'under 10% of full model' parameters is configuration-dependent—Table 4 shows ST-LoRA trainable parameters exceeding the base model in several rows—so a fair efficiency comparison should report trainable parameters against a fixed base-model definition and equal ensemble size.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes ST-LoRA, a parameter-efficient ensemble method that combines LoRA adapters with snapshot ensembling for uncertainty-aware semantic segmentation in agriculture. A frozen pretrained backbone is shared across ensemble members, and members differ only in low-rank adapters collected from a single cosine-annealing trajectory. The method is evaluated on GrowliFlower-L (cauliflower) and BUP20 (sweet pepper) with SegFormer-B2/B4 and Mask2Former-Swin-B, comparing against a full-rank snapshot ensemble (FRE), MC Dropout, and DDU. The authors report competitive mIoU and ECE, calibration robustness under distribution shift, OoD detection, and edge-hardware latency gains, and they claim that feed-forward layers are the critical LoRA target for dense prediction. Extensive ablations cover rank, scaling, dropout, target components, and architectural choices with 5 seeds.

Significance. If the central claims held, ST-LoRA would be a practically useful method for uncertainty quantification in resource-constrained agricultural vision, and the layer-target finding would extend the LoRA-ensemble literature from image classification/NLP to dense prediction. The paper has clear strengths: code is provided, standard metrics are used, results are averaged over 5 seeds, ablations are systematic, and deployment latency is measured on real edge hardware. However, two load-bearing claims are currently overstated: the 'under 10% trainable parameters' claim is contradicted by the paper's own tables, and 'matches full-rank ensembles' is only established against a snapshot ensemble, not the standard independently trained Deep Ensemble. These issues are fixable but require either new experiments or substantial qualification.

major comments (3)
  1. [Abstract; §4.3.1, Table 4; §6.3] The abstract and §6.3 claim that trainable parameters are reduced to 'under 10% of the full model.' Table 4 contradicts this: ST-LoRA has 70.84M vs 61.37M base (Sgf-B4), 29.85M vs 24.72M (Sgf-B2), and 116M vs 108.77M (M2F), i.e., 107–121% of the base model's trainable parameters. Table 13 lists the baseline ST-LoRA configuration as 11.19% trainable parameters, already above 10%. Please reconcile these numbers and either define 'trainable parameters' explicitly (LoRA adapters only vs. all updated weights) or correct the headline efficiency claim.
  2. [§5.1, Eq. (10), footnote 1; Tables 3, 9, 11] The FRE baseline is a snapshot ensemble of checkpoints from one cosine-annealing trajectory, not the standard Deep Ensemble of M independently initialized and trained models (Lakshminarayanan et al. 2017). No independently trained ensemble baseline appears in Tables 3, 9, or 11. Therefore the abstract's claim that ST-LoRA 'matches or exceeds full-rank ensembles' is established only for full-rank snapshot ensembles. Since snapshot members share a trajectory, they are less diverse than deep ensembles, and the trade-off against the strongest full-rank baseline remains untested. Please either add a Deep Ensemble baseline or qualify all 'full-rank ensemble' claims as 'full-rank snapshot ensemble.'
  3. [§1, §7, §5.4] The paper repeatedly claims that ST-LoRA 'reduces training time by 1/M.' Because FRE is also collected from a single training run (§5.1 footnote 1), both methods incur the same one-trajectory training cost; the 1/M factor applies only relative to training M independent full models. This is another manifestation of the mismatch between the claimed baseline (Deep Ensemble) and the actual baseline (snapshot ensemble). Please clarify the reference point for all training-time and efficiency claims.
minor comments (6)
  1. [§5.1, Table 3 (BUP20/SgF-B2 row)] The ΔECE for ST-LoRA is printed as '+25', but 0.0021 vs. the baseline 0.0028 is a 25% decrease (−25%). The accompanying text also says ST-LoRA achieves the best mIoU among all methods and a 'marginal ECE increase relative to FRE'; in this row FRE mIoU is 70.0 vs. 69.3, and ST-LoRA ECE is lower than FRE. Please correct the sign and the interpretation.
  2. [Table 4] The M2F FLOPs entry is written '229,72' (comma decimal). Also clarify whether FLOPs are reported per ensemble member or as the total for the full ensemble; the values in Table 4 appear to be totals, but the text calls them 'comparable' without this distinction.
  3. [Table 11] The two 'AUROC↑' columns have identical headers. Label them as 'Image-level AUROC' and 'Pixel-level AUROC' to avoid ambiguity.
  4. [§4.1 vs. §5.4; Table 2] §4.1 says 'No augmentation has been used,' and Table 2 reports 'Augmentation None,' but §5.3 and §5.4 use a heterogeneous augmentation pipeline shared by all methods. Specify which augmentation setting applies to each experimental block, since this affects the diversity claims.
  5. [Appendix / Tables 4, 8] The ensemble size is stated as 5 in Table 2 and §5.1, but Tables 4, 8, and the appendix use 4 checkpoints ('retain the last four checkpoints'). Clarify whether the first snapshot is always discarded and report the effective M consistently.
  6. [References] Several references are incomplete, including Hu et al. (LoRA), He et al. (unified view of PEFT), Onal et al., and Halbheer et al. Missing years/venues should be added.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ST-LoRA is an empirical evaluation; its uncertainty estimates are computed from held-out test predictions, and its baselines are independently run.

full rationale

The paper's derivation chain is empirical rather than formal. ST-LoRA's predictive posterior (Eq. 10) is a standard uniform mixture over adapters; the uncertainty measures (Eqs. 4-7) are applied to model outputs and are not fitted to the quantities they are claimed to predict. Hyperparameters are ablated across separate runs and final configurations are evaluated on held-out test splits (Tables 2 and 6), so no parameter is fit to the target metric and then reported as a prediction. The FRE baseline is explicitly defined as a snapshot ensemble of the fully fine-tuned model (footnote 1, page 2; Eq. 10 context), and the comparison against it is a direct experimental measurement, not a consequence of the definition. The abstract's 'under 10%' parameter claim is inconsistent with Table 4, and the phrase 'full-rank ensembles' could be read as referring to the standard Deep Ensemble rather than the paper's FRE; these are correctness/communication concerns, not circular reasoning. Self-citations (e.g., Farag et al. 2025 for agricultural UQ gaps; Kierdorf et al. for GrowliFlower-L) are contextual and not load-bearing for any derived result. The paper's own limitations section explicitly acknowledges hyperparameter sensitivity, residual miscalibration, data/task dependency, and deployment overhead, further supporting that the empirical claims are not presented as forced by construction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper's claims rest on standard LoRA and snapshot ensembling assumptions, on the representativeness of the two datasets, and on the validity of synthetic corruptions as shift. No new physical entities or ad hoc assumptions are introduced beyond standard hyperparameters.

free parameters (4)
  • LoRA rank r
    Hyperparameter set to 8 in the main ID comparison and up to 32 in ablations; chosen by hand based on validation, not derived from theory.
  • LoRA scaling alpha
    Hyperparameter ablated from 2 to 128; alpha/r controls the adapter update magnitude and is treated as a deployment-time calibration knob.
  • LoRA dropout p
    Ablated over 0.1-0.3; chosen per dataset based on calibration and robustness experiments.
  • Ensemble size M = 4 or 5
    Number of snapshots used per ensemble; set to 5 in the main experiments and 4 in the later comparison with baselines.
assumptions (4)
  • domain assumption Cosine-annealing snapshot ensembling produces diverse model states that approximate samples from the posterior.
    Relies on Huang et al. (2017); the paper does not independently justify that the saved checkpoints are diverse enough for UQ.
  • domain assumption Low-rank adaptation preserves task-relevant information, following the intrinsic-dimensionality argument of Aghajanyan et al. and the correlation analysis of Hu et al.
    Used to claim LoRA members can form meaningful uncertainty estimates; drawn from prior literature, not re-verified here.
  • domain assumption Synthetic corruptions and a different crop (cauliflower vs pepper) are valid proxies for deployment-time distribution shift and far OoD inputs.
    The entire robustness evaluation treats these as realistic, but no real-world shift data is tested.
  • domain assumption The two datasets (GrowliFlower-L and BUP20) and the two transformer architectures are representative enough to support a general practitioner's guide.
    The guide is intended to transfer beyond these datasets, but evidence is limited to two agricultural crops and two backbones.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ST-LoRA: Single Trajectory LoRA Ensemble for Uncertainty Aware Agricultural Segmentation." pith.science (2026). https://pith.science/paper/5NABIJ2N

@misc{pith2026260801530,
  author       = {Pith},
  title        = {Pith review of: ST-LoRA: Single Trajectory LoRA Ensemble for Uncertainty Aware Agricultural Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5NABIJ2N}},
  note         = {Machine review of arXiv:2608.01530}
}
read the original abstract

Reliable decision-support in digital agriculture requires accurate predictions and well-calibrated uncertainty estimates, particularly for dense prediction tasks such as semantic segmentation. Ensemble methods provide strong uncertainty quantification, but their computational and memory demands limit practical use, while single-model approximations often trade off uncertainty quality for efficiency. We propose ST-LoRA, a parameter-efficient ensemble framework that builds diverse ensemble members from a single training trajectory by combining Low-Rank Adaptation (LoRA) with snapshot ensembling. Each member shares a frozen pretrained backbone and differs only in lightweight low-rank adapters, reducing trainable parameters to under 10% of the full model while preserving ensemble diversity. We evaluate across two agricultural datasets - GrowliFlower-L (cauliflower, open field) and BUP20 (sweet pepper, glasshouse) - using SegFormer and Mask2Former, covering in-distribution performance, calibration under distribution shift, and out-of-distribution detection. Ablations show feed-forward layers, not attention layers, are the critical LoRA target for dense prediction, contrary to the attention-only convention from language models. ST-LoRA matches or exceeds full-rank ensembles in segmentation accuracy and calibration across both datasets and architectures, while substantially reducing training time, inference latency, memory footprint, and storage requirements. Against efficient baselines - Snapshot Ensemble, MC Dropout, and Deep Deterministic Uncertainty - ST-LoRA consistently matches or outperforms them in image/pixel-level OoD detection, calibration stability under shift, and cross-seed variance, with far fewer parameters and lower compute. These results show LoRA-efficient ensemble adaptation is a highly effective, practical approach for uncertainty-aware agricultural vision systems.

Figures

Figures reproduced from arXiv: 2608.01530 by the authors.

Figure 1
Figure 1. ST-LoRA for Efficient Uncertainty Quantification. The upper part illustrates the fine-tuning process of the semantic segmentation model, where both the input image and ground truth (for simplicity we overlay the ground truth mask over the input image) are provided. Only ST-LoRA modules are fine-tuned, while the rest of the model remains frozen. During inference, an ensemble of ST-LoRA-adapters is applied to the inpu… view at source ↗
Figure 2
Figure 2. LoRA Ensemble Visual Interpretation. A) a DL model once fully trained, the hypothesis space collapses (represented by the gray shade) ℍ to a single deterministic parameter 𝜃0 leaving no ability to explore the hypothesis space without retraining. B), the standard ensembling approach is shown, where multiple models are independently updated via full-rank weight shifts Δ𝜃 based on the downstream task. C) by sampling or… view at source ↗
Figure 3
Figure 3. Dataset samples. Left: Sweet pepper (UGV) with semantic segmentation masks for different pepper classes. Right: Cauliflower (UAV) with instance masks; we convert these to binary semantic masks (plant vs. soil) for our experiments. Farag et al.: Preprint submitted to Elsevier Page 7 of 28 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Cosine annealing learning rate schedule used for snapshot ensemble construction. Each cycle drives the model toward a distinct local minimum, and a checkpoint is saved at the end of each cycle (marked by vertical dashed lines). The resulting 𝑀 snapshots form the ensemb…
Figure 5
Figure 5. Figure 5: GrowliFlower-L reliability diagram for ID calibration assessment. As shown in the figure, we compare four different settings for both model (e.g., FRE, ST-LoRA, single LoRA adapter, and Baseline model). The ensemble of ST-LoRA enhances calibration by uplifting the plot…
Figure 6
Figure 6. Figure 6: Ensemble robustness under covariate shift. Homogeneous vs. heterogeneous ensemble variants for ST-LoRA and FRE on GrowliFlower-L across 80 shift conditions. suggesting that a low-dimensional adaptation subspace lacks sufficient capacity to maintain well-calibrated pred…
Figure 7
Figure 7. Figure 7: mIoU under distribution shift as a function of LoRA rank 𝑟, evaluated across seven corruption types at three severity levels. Each curve corresponds to a distinct shift type [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: ECE under distribution shift as a function of LoRA rank 𝑟, evaluated across seven corruption types at three severity levels. Each curve corresponds to a distinct shift type. Target components. Ranking configurations by mIoU and ECE robustness under distribution shift r…
Figure 9
Figure 9. Figure 9: Robustness under distribution shift comparison. ST-LoRA demonstrates the best performance against different kinds of shifts. ↑: higher is better; ↓: lower is better. Represented the mean of five different seeds. [0.9, 1.1]). Geometric corruptions include lateral transl…
Figure 10
Figure 10. Figure 10: Image-level OoD score distributions across methods. Each panel shows the density of predictive entropy (left) and mutual information/Density (right), dashed vertical lines indicate the mean of each distribution. We decide to present the best performance for each metho…
Figure 11
Figure 11. Figure 11: Hyperparameters robustness against distribution shift measured by mIoU metric. Farag et al.: Preprint submitted to Elsevier Page 31 of 28 [PITH_FULL_IMAGE:figures/full_fig_p032_11.png]
Figure 12
Figure 12. Figure 12: Hyperparameters robustness against distribution shift measured by ECE metric. Farag et al.: Preprint submitted to Elsevier Page 32 of 28 [PITH_FULL_IMAGE:figures/full_fig_p033_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

202 extracted references · 40 canonical work pages

  1. [1]

    GrowliFlower: An image time-series dataset for GROWth analysis of cauLIFLOWER , year =

    Kierdorf, Jana and Junker-Frohn, Laura Verena and Delaney, Mike and Olave, Mariele Donoso and Burkart, Andreas and Jaenicke, Hannah and Muller, Onno and Rascher, Uwe and Roscher, Ribana , journal =. GrowliFlower: An image time-series dataset for GROWth analysis of cauLIFLOWER , year =. doi:https://doi.org/10.1002/rob.22122 , groups =

    work page doi:10.1002/rob.22122

  2. [2]

    Rajendra Acharya and Vladimir Makarenkov and Saeid Nahavandi , journal =

    Moloud Abdar and Farhad Pourpanah and Sadiq Hussain and Dana Rezazadegan and Li Liu and Mohammad Ghavamzadeh and Paul Fieguth and Xiaochun Cao and Abbas Khosravi and U. Rajendra Acharya and Vladimir Makarenkov and Saeid Nahavandi , journal =. A review of uncertainty quantification in deep learning: Techniques, applications and challenges , year =. doi:htt...

    doi:10.1016/j.inffus.2021.05.008 2021

  3. [3]

    Machine Learning , title =

    Eyke H. Machine Learning , title =. 2019 , pages =

    2019

  4. [4]

    Deep Ensembles from a Bayesian Perspective

    Lara Hoffmann and Clemens Elster , journal =. Deep Ensembles from a Bayesian Perspective , year =. arXiv , bibsource =:2105.13283 , groups =

    work page Pith review arXiv

  5. [5]

    ArXiv , title =

    Cornelia Gruber and Patrick Oliver Schenk and Malte Schierholz and Frauke Kreuter and G. ArXiv , title =. 2023 , volume =

    2023

  6. [6]

    Weinberger , booktitle =

    Chuan Guo and Geoff Pleiss and Yu Sun and Kilian Q. Weinberger , booktitle =. On Calibration of Modern Neural Networks , year =

  7. [7]

    A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification , year =

    Anastasios Nikolas Angelopoulos and Stephen Bates , journal =. A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification , year =

  8. [8]

    Distribution-free prediction bands for non-parametric regression , year =

    Jing Lei and Larry Wasserman , journal =. Distribution-free prediction bands for non-parametric regression , year =

Show all 202 references
  1. [9]

    Proceedings of the Asian Conference on Machine Learning , pages =

    Conditional Validity of Inductive Conformal Predictors , author =. Proceedings of the Asian Conference on Machine Learning , pages =. 2012 , editor =

    2012

  2. [10]

    Inductive Confidence Machines for Regression , year =

    Papadopoulos, Harris and Proedrou, Kostas and Vovk, Volodya and Gammerman, Alex , booktitle =. Inductive Confidence Machines for Regression , year =

  3. [11]

    Algorithmic Learning in a Random World , year =

    Vladimir Vovk and Alexander Gammerman and Glenn Shafer , publisher =. Algorithmic Learning in a Random World , year =. doi:10.1007/978-3-031-06649-8 , groups =

    doi

  4. [12]

    International Conference on Machine Learning , year=

    Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning , author=. International Conference on Machine Learning , year=

  5. [13]

    Improving wheat yield estimates using data augmentation models and remotely sensed biophysical indices within deep neural networks in the Guanzhong Plain, PR China , year =

    Jingqi Zhang and Huiren Tian and Pengxin Wang and Kevin Tansey and Shuyu Zhang and Hongmei Li , journal =. Improving wheat yield estimates using data augmentation models and remotely sensed biophysical indices within deep neural networks in the Guanzhong Plain, PR China , year...

    2021

  6. [14]

    Maize seedling detection under different growth stages and complex field environments based on an improved Faster R–CNN , year =

    Longzhe Quan and Huaiqu Feng and Yingjie Lv and Qi Wang and Chuanbin Zhang and Jingguo Liu and Zongyang Yuan , journal =. Maize seedling detection under different growth stages and complex field environments based on an improved Faster R–CNN , year =. doi:https://doi.org/10.10...

    2019 doi

  7. [15]

    Deep learning based banana plant detection and counting using high-resolution red-green-blue (RGB) images collected from unmanned aerial vehicle (UAV) , year =

    Bipul Neupane and Teerayut Horanont and Nguyen Duy Hung , journal =. Deep learning based banana plant detection and counting using high-resolution red-green-blue (RGB) images collected from unmanned aerial vehicle (UAV) , year =

  8. [16]

    Ferentinos , journal =

    Konstantinos P. Ferentinos , journal =. Deep learning models for plant disease detection and diagnosis , year =. doi:https://doi.org/10.1016/j.compag.2018.01.009 , groups =

    2018 doi

  9. [17]

    Deep machine learning provides state-of-the-art performance in image-based plant phenotyping , year =

    Pound, Michael P and Atkinson, Jonathan A and Townsend, Alexandra J and Wilson, Michael H and Griffiths, Marcus and Jackson, Aaron S and Bulat, Adrian and Tzimiropoulos, Georgios and Wells, Darren M and Murchie, Erik H and others , journal =. Deep machine learning provides sta...

  10. [18]

    and Cunha, António , journal =

    Teixeira, Igor and Morais, Raul and Sousa, Joaquim J. and Cunha, António , journal =. Deep Learning Models for the Classification of Crops in Aerial Imagery: A Review , year =. doi:10.3390/agriculture13050965 , groups =

    doi

  11. [19]

    DAGM German Conference on Pattern Recognition , title =

    Drees, Lukas and Weber, Immanuel and Ru. DAGM German Conference on Pattern Recognition , title =. 2022 , organization =

    2022

  12. [20]

    2023 , issue_date =

    Gawlikowski, Jakob and Tassi, Cedrique Rovile Njieutcheu and Ali, Mohsin and Lee, Jongseok and Humt, Matthias and Feng, Jianxiang and Kruspe, Anna and Triebel, Rudolph and Jung, Peter and Roscher, Ribana and Shahzad, Muhammad and Yang, Wen and Bamler, Richard and Zhu, Xiao Xia...

    2023 doi

  13. [21]

    A Survey on Bayesian Deep Learning , year =

    Wang, Hao and Yeung, Dit-Yan , journal =. A Survey on Bayesian Deep Learning , year =. doi:10.1145/3409383 , groups =

    doi

  14. [22]

    Simple and Scalable Predictive Uncertainty Estimation Using Deep Ensembles , year =

    Lakshminarayanan, Balaji and Pritzel, Alexander and Blundell, Charles , booktitle =. Simple and Scalable Predictive Uncertainty Estimation Using Deep Ensembles , year =

  15. [23]

    Algorithmic Learning in a Random World , year =

    Vladimir Vovk, Alexander Gammerman, Glenn Shafer , publisher =. Algorithmic Learning in a Random World , year =. doi:https://doi.org/10.1007/b106715 , groups =

    doi

  16. [24]

    Flipout: Efficient Pseudo-Independent Weight Perturbations on Mini-Batches , year =

    Yeming Wen and Paul Vicol and Jimmy Ba and Dustin Tran and Roger Baker Grosse , journal =. Flipout: Efficient Pseudo-Independent Weight Perturbations on Mini-Batches , year =

  17. [25]

    Proceedings of the 37th International Conference on Neural Information Processing Systems , articleno =

    Huang, Ziyi and Lam, Henry and Zhang, Haofeng , title =. Proceedings of the 37th International Conference on Neural Information Processing Systems , articleno =. 2024 , publisher =

    2024

  18. [26]

    International Conference on Machine Learning , year=

    Uncertainty Estimation Using a Single Deep Deterministic Neural Network , author=. International Conference on Machine Learning , year=

  19. [27]

    and Gal, Yarin , booktitle =

    Mukhoti, Jishnu and Kirsch, Andreas and van Amersfoort, Joost and Torr, Philip H.S. and Gal, Yarin , booktitle =. Deep Deterministic Uncertainty: A New Simple Baseline , year =

  20. [28]

    On the Practicality of Deterministic Epistemic Uncertainty , year =

    Janis Postels and Mattia Segu and Tao Sun and Luc Van Gool and Fisher Yu and Federico Tombari , booktitle =. On the Practicality of Deterministic Epistemic Uncertainty , year =

  21. [29]

    Proceedings of the 31st International Conference on Neural Information Processing Systems , pages =

    Kendall, Alex and Gal, Yarin , title =. Proceedings of the 31st International Conference on Neural Information Processing Systems , pages =. 2017 , isbn =

    2017

  22. [30]

    Proceedings of the 40th International Conference on Machine Learning , articleno =

    Kirchhof, Michael and Kasneci, Enkelejda and Oh, Seong Joon , title =. Proceedings of the 40th International Conference on Machine Learning , articleno =. 2023 , publisher =

    2023

  23. [31]

    Probabilistic machine learning and artificial intelligence , year =

    Ghahramani, Zoubin , journal =. Probabilistic machine learning and artificial intelligence , year =. doi:10.1038/nature14541 , groups =

    doi

  24. [32]

    Optimal zinc level and uncertainty quantification in agricultural soils via visible near-infrared reflectance and soil chemical properties , year =

    Prince Chapman Agyeman and Ndiye Michael Kebonye and Vahid Khosravi and John Kingsley and Luboš Borůvka and Radim Vašát and Charles Mario Boateng , journal =. Optimal zinc level and uncertainty quantification in agricultural soils via visible near-infrared reflectance and soil...

    2022

  25. [35]

    Inductive Conformal Prediction for Harvest-Readiness Classification of Cauliflower Plants: A Comparative Study of Uncertainty Quantification Methods , year =

    Farag, Mohamed and Kierdorf, Jana and Roscher, Ribana , booktitle =. Inductive Conformal Prediction for Harvest-Readiness Classification of Cauliflower Plants: A Comparative Study of Uncertainty Quantification Methods , year =. doi:10.1109/ICCVW60793.2023.00072 , groups =

    2023

  26. [36]

    and Dolias, Georgios and Berruto, Remigio and Kateris, Dimitrios and Bochtis, Dionysis , journal =

    Benos, Lefteris and Tagarakis, Aristotelis C. and Dolias, Georgios and Berruto, Remigio and Kateris, Dimitrios and Bochtis, Dionysis , journal =. Machine Learning in Agriculture: A Comprehensive Updated Review , year =. doi:10.3390/s21113758 , groups =

    doi

  27. [37]

    Ramkteke , journal =

    Vishal Meshram and Kailas Patil and Vidula Meshram and Dinesh Hanchate and S.D. Ramkteke , journal =. Machine learning in agriculture domain: A state-of-art survey , year =. doi:https://doi.org/10.1016/j.ailsci.2021.100010 , groups =

    2021

  28. [38]

    Uncertainty Quantification of Soil Organic Carbon Estimation from Remote Sensing Data with Conformal Prediction , year =

    Kakhani, Nafiseh and Alamdar, Setareh and Kebonye, Ndiye Michael and Amani, Meisam and Scholten, Thomas , journal =. Uncertainty Quantification of Soil Organic Carbon Estimation from Remote Sensing Data with Conformal Prediction , year =. doi:10.3390/rs16030438 , groups =

    doi

  29. [39]

    Evaluating Bayesian Deep Learning Methods for Semantic Segmentation , year =

    Jishnu Mukhoti and Yarin Gal , journal =. Evaluating Bayesian Deep Learning Methods for Semantic Segmentation , year =

  30. [40]

    and Schaaf, Gabriel and Gall, Juergen , journal =

    Yi, Jinhui and Krusenbaum, Lukas and Unger, Paula and Hüging, Hubert and Seidel, Sabine J. and Schaaf, Gabriel and Gall, Juergen , journal =. Deep Learning for Non-Invasive Diagnosis of Nutrient Deficiencies in Sugar Beet Using RGB Images , year =. doi:10.3390/s20205893 , groups =

    doi

  31. [41]

    Vapnik, V. N. and Chervonenkis, A. Ya. , pages =. On the Uniform Convergence of Relative Frequencies of Events to Their Probabilities , year =. Measures of Complexity , doi =

  32. [44]

    References , year =

    Cristianini, Nello and Shawe-Taylor, John , pages =. References , year =. An Introduction to Support Vector Machines and Other Kernel-based Learning Methods , place =

  33. [45]

    Littlestone, Nick and Warmuth, Manfred K , title =

  34. [46]

    , booktitle =

    McAllester, David A. , booktitle =. Some PAC-Bayesian theorems , year =. doi:10.1145/279943.279989 , groups =

  35. [47]

    International Conference on Learning Representations , year=

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. International Conference on Learning Representations , year=

  36. [48]

    Deep Residual Learning for Image Recognition , year=

    He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian , booktitle=. Deep Residual Learning for Image Recognition , year=

  37. [49]

    and Ba, Jimmy , title =

    Kingma, Diederik P. and Ba, Jimmy , title =. 2014 , booktitle=

    2014

  38. [50]

    , journal =

    Krizhevsky, Alex and Sutskever, Ilya and Hinton, Geoffrey E. , journal =. ImageNet classification with deep convolutional neural networks , year =. doi:10.1145/3065386 , groups =

    doi

  39. [51]

    , journal =

    Gneiting, Tilmann and Balabdaoui, Fadoua and Raftery, Adrian E. , journal =. Probabilistic forecasts, calibration and sharpness , year =. doi:https://doi.org/10.1111/j.1467-9868.2007.00587.x , eprint =

    2007

  40. [52]

    Sculley and Sebastian Nowozin and Joshua V

    Yaniv Ovadia and Emily Fertig and Jie Jessie Ren and Zachary Nado and D. Sculley and Sebastian Nowozin and Joshua V. Dillon and Balaji Lakshminarayanan and Jasper Snoek , booktitle =. Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift...

  41. [53]

    FAO and IFAD and UNICEF and WFP and WHO , year =. The

  42. [54]

    Counting of grapevine berries in images via semantic segmentation using convolutional neural networks , year =

    Laura Zabawa and Anna Kicherer and Lasse Klingbeil and Reinhard Töpfer and Heiner Kuhlmann and Ribana Roscher , journal =. Counting of grapevine berries in images via semantic segmentation using convolutional neural networks , year =. doi:https://doi.org/10.1016/j.isprsjprs.20...

    2020 doi

  43. [55]

    2021 , doi =

    Shaker, Mohammad Hossein and Hüllermeier, Eyke , title =. 2021 , doi =

    2021

  44. [56]

    Conference on Uncertainty in Artificial Intelligence , title =

    Eyke H. Conference on Uncertainty in Artificial Intelligence , title =. 2022 , groups =

    2022

  45. [57]

    2023 , note=

    Salem Lahlou and Moksh Jain and Hadi Nekoei and Victor I Butoi and Paul Bertin and Jarrid Rector-Brooks and Maksym Korablyov and Yoshua Bengio , journal=. 2023 , note=

    2023

  46. [58]

    A Unified Survey on Anomaly, Novelty, Open-Set, and Out-of-Distribution Detection: Solutions and Future Challenges , author=. Trans. Mach. Learn. Res. , year=

  47. [59]

    Frontiers in Psychology , VOLUME=

    De Groot, Kristel and Thurik, Roy , TITLE=. Frontiers in Psychology , VOLUME=. 2018 , DOI=

    2018

  48. [60]

    , year =

    Knight, Frank H. , year =. Risk,

  49. [61]

    The Twelfth International Conference on Learning Representations , year=

    Conformal Risk Control , author=. The Twelfth International Conference on Learning Representations , year=

  50. [62]

    Shannon, C. E. , title =. Bell System Technical Journal , volume =. doi:https://doi.org/10.1002/j.1538-7305.1948.tb01338.x , eprint =

    1948

  51. [63]

    2012 , issn =

    A unifying view on dataset shift in classification , journal =. 2012 , issn =. doi:https://doi.org/10.1016/j.patcog.2011.06.019 , author =

    2012 doi

  52. [64]

    and Paul, Madhuri R

    Drees, Lukas and Demie, Dereje T. and Paul, Madhuri R. and Leonhardt, Johannes and Seidel, Sabine J. and Döring, Thomas F. and Roscher, Ribana , journal =. Data-driven crop growth simulation on time-varying generated images using multi-conditional generative adversarial networ...

    doi

  53. [65]

    2024 , issn =

    How can we quantify, explain, and apply the uncertainty of complex soil maps predicted with neural networks? , journal =. 2024 , issn =. doi:https://doi.org/10.1016/j.scitotenv.2024.173720 , author =

    2024

  54. [66]

    The Nature of Statistical Learning Theory , year =

    Vladimir Vapnik , editor =. The Nature of Statistical Learning Theory , year =

  55. [67]

    Valiant, L. G. , title =. 1984 , issue_date =. doi:10.1145/1968.1972 , journal =

    1984

  56. [68]

    Machine Learning , author =

    Sample. Machine Learning , author =. 1995 , keywords =. doi:10.1023/A:1022660318680 , language =

    1995 doi

  57. [69]

    An Introduction to Support Vector Machines and Other Kernel-based Learning Methods , publisher=

    Cristianini, Nello and Shawe-Taylor, John , year=. An Introduction to Support Vector Machines and Other Kernel-based Learning Methods , publisher=

  58. [70]

    Relating Data Compression and Learnability , author=

  59. [71]

    Vapnik, V. N. and Chervonenkis, A. Ya. , title =. Springer International Publishing , doi =. 2015 , address =

    2015

  60. [72]

    2014 , abstract =

    Wood, Michael , month = feb, title =. 2014 , abstract =. doi:10.2139/ssrn.2393927 , file =

    2014 doi

  61. [73]

    Hinton and Nitish Srivastava and Alex Krizhevsky and Ilya Sutskever and Ruslan Salakhutdinov , title =

    Geoffrey E. Hinton and Nitish Srivastava and Alex Krizhevsky and Ilya Sutskever and Ruslan Salakhutdinov , title =. CoRR , volume =. 2012 , eprinttype =. 1207.0580 , timestamp =

    2012 arXiv

  62. [74]

    Philosophical Transactions of the Royal Society of London

    On the Problem of the Most Efficient Tests of Statistical Hypotheses , author=. Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character , volume=. 1933 , publisher=

    1933

  63. [75]

    Neural Comput

    Rosasco, Lorenzo and De Vito, Ernesto and Caponnetto, Andrea and Piana, Michele and Verri, Alessandro , title =. Neural Comput. , month =. 2004 , issue_date =. doi:10.1162/089976604773135104 , abstract =

    2004 doi

  64. [76]

    Biometrika , volume=

    A new measure of rank correlation , author=. Biometrika , volume=. 1938 , publisher=

    1938

  65. [77]

    The American Journal of Psychology , volume=

    The proof and measurement of association between two things , author=. The American Journal of Psychology , volume=. 1904 , publisher=

    1904

  66. [78]

    Proceedings of the IEEE , volume=

    Gradient-based learning applied to document recognition , author=. Proceedings of the IEEE , volume=. 1998 , publisher=

    1998

  67. [79]

    2009 , institution=

    Learning multiple layers of features from tiny images , author=. 2009 , institution=

    2009

  68. [80]

    IEEE Transactions on Systems Science and Cybernetics , volume=

    Visual feature extraction by a multilayered network of analog threshold elements , author=. IEEE Transactions on Systems Science and Cybernetics , volume=. 1969 , publisher=

    1969

  69. [81]

    Proceedings of the 34th International Conference on Machine Learning-Volume 70 , pages=

    On Calibration of Modern Neural Networks , author=. Proceedings of the 34th International Conference on Machine Learning-Volume 70 , pages=. 2017 , organization=

    2017

  70. [82]

    Proceedings of the 32nd International Conference on Machine Learning , pages =

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , author =. Proceedings of the 32nd International Conference on Machine Learning , pages =. 2015 , editor =

    2015

  71. [83]

    Hedging Predictions in Machine Learning , year =

    Gammerman, Alexander and Vovk, Vladimir , journal =. Hedging Predictions in Machine Learning , year =. doi:10.1093/comjnl/bxl065 , publisher =

    doi

  72. [84]

    Machine Learning , author =

    Conformal predictions for probabilistically robust scalable machine learning classification , volume =. Machine Learning , author =. 2024 , keywords =. doi:10.1007/s10994-024-06571-6 , abstract =

    2024 doi

  73. [85]

    ICML 2024 Workshop on Structured Probabilistic Inference

    Conformalized Credal Set Predictors , author=. ICML 2024 Workshop on Structured Probabilistic Inference

    2024

  74. [86]

    ArXiv , year=

    Conformal prediction under ambiguous ground truth , author=. ArXiv , year=

  75. [87]

    Annals of Mathematics and Artificial Intelligence , year=

    Cross-conformal predictors , author=. Annals of Mathematics and Artificial Intelligence , year=

  76. [88]

    The Annals of Statistics , year=

    Conformal prediction beyond exchangeability , author=. The Annals of Statistics , year=

  77. [89]

    ArXiv , year=

    Benchmarking Uncertainty Disentanglement: Specialized Uncertainties for Specialized Tasks , author=. ArXiv , year=

  78. [90]

    ArXiv , year=

    How disentangled are your classification uncertainties? , author=. ArXiv , year=

  79. [91]

    ArXiv , year=

    Quantifying Aleatoric and Epistemic Uncertainty with Proper Scoring Rules , author=. ArXiv , year=

  80. [92]

    ArXiv , year=

    Predictive Uncertainty Quantification via Risk Decompositions for Strictly Proper Scoring Rules , author=. ArXiv , year=

  81. [93]

    Status of the World’s Soil Resources , institution =

  82. [94]

    Sustainable Plant Nutrition , publisher =

    Mustafa, Ghulam and Hayat, Naveed and Alotaibi, Bader Alhafi , title =. Sustainable Plant Nutrition , publisher =. 2023 , pages =. doi:10.1016/B978-0-443-18675-2.00019-5 , url =

    2023 doi

  83. [95]

    , title =

    Lowe, N.M. , title =. Proceedings of the Nutrition Society , year =. doi:10.1017/S0029665121000902 , note =

    doi

  84. [96]

    Journal of Geophysical Research: Biogeosciences , volume =

    Wang, Hui and Wellmann, Florian and Zhang, Tianqi and Schaaf, Alexander and Kanig, Robin Maximilian and Verweij, Elizabeth and von Hebel, Christian and van der Kruk, Jan , title =. Journal of Geophysical Research: Biogeosciences , volume =. doi:https://doi.org/10.1029/2019JG00...

    doi

  85. [97]

    and Chen, N

    Xu, L. and Chen, N. and Yang, C. and Yu, H. and Chen, Z. , TITLE =. Hydrology and Earth System Sciences , VOLUME =. 2022 , NUMBER =

    2022

  86. [98]

    and Triggs, Christopher M

    Meenken, Esther D. and Triggs, Christopher M. and Brown, Hamish E. and Sinton, Sarah and Bryant, Jeremy R. and Noble, Alasdair D.L. and Espig, Martin and Sharifi, Mostafa and Wheeler, David M. , title =. Agronomy Journal , volume =. doi:https://doi.org/10.1002/agj2.20659 , url...

    doi

  87. [99]

    and Jenkins, Eleanor W

    Chrispell, John C. and Jenkins, Eleanor W. and Kavanagh, Kathleen R. and Parno, Matthew D. , TITLE =. Modelling , VOLUME =. 2021 , NUMBER =

    2021

  88. [100]

    Concurrency and Computation: Practice and Experience , volume =

    Rajamanickam, Jayalakshmi and Mani, Savitha Devi , title =. Concurrency and Computation: Practice and Experience , volume =. doi:https://doi.org/10.1002/cpe.6460 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1002/cpe.6460 , year =

    doi

  89. [101]

    Hernández and Juan L

    S. Hernández and Juan L. López , keywords =. Uncertainty quantification for plant disease detection using Bayesian deep learning , journal =. 2020 , issn =. doi:https://doi.org/10.1016/j.asoc.2020.106597 , url =

    2020

  90. [102]

    Padarian and B

    J. Padarian and B. Minasny and A.B. McBratney , keywords =. Assessing the uncertainty of deep learning soil spectral models using Monte Carlo dropout , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.geoderma.2022.116063 , url =

    2022

  91. [103]

    Simis and Daniela Gurlin and Moritz K

    Mortimer Werther and Daniel Odermatt and Stefan G.H. Simis and Daniela Gurlin and Moritz K. Lehmann and Tiit Kutser and Remika Gupana and Adam Varley and Peter D. Hunter and Andrew N. Tyler and Evangelos Spyrakos , keywords =. A Bayesian approach for remote sensing of chloroph...

    2022

  92. [104]

    Quantifying uncertainty in soil moisture retrieval using a Bayesian neural network framework , journal =

    Yan Li and Songhua Yan and Jianya Gong , keywords =. Quantifying uncertainty in soil moisture retrieval using a Bayesian neural network framework , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.compag.2023.108414 , url =

    2023

  93. [105]

    2025 , url=

    Epistemic Uncertainty in Conformal Scores: A Unified Approach , author=. 2025 , url=

    2025

  94. [106]

    ArXiv , year=

    Out-of-Distribution Detection Should Use Conformal Prediction (and Vice-versa?) , author=. ArXiv , year=

  95. [107]

    Advances in Neural Information Processing Systems 33 (NeurIPS 2020) , year=

    Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness , author=. Advances in Neural Information Processing Systems 33 (NeurIPS 2020) , year=

    2020

  96. [108]

    Proceedings of the IEEE International Conference on Neural Networks (ICNN) , pages=

    Estimating the Mean and Variance of the Target Probability Distribution in Neural Network Training , author=. Proceedings of the IEEE International Conference on Neural Networks (ICNN) , pages=

  97. [109]

    Econometrica , volume=

    Regression Quantiles , author=. Econometrica , volume=. 1978 , publisher=

    1978

  98. [110]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Do Deep Generative Models Know What They Don't Know? , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  99. [111]

    ICML 2024 Workshop on Structured Probabilistic Inference

    Quantifying Aleatoric and Epistemic Uncertainty: A Credal Approach , author=. ICML 2024 Workshop on Structured Probabilistic Inference. 2024 , url=

    2024

  100. [112]

    International Conference on Machine Learning , year=

    Position: Bayesian Deep Learning is Needed in the Age of Large-Scale AI , author=. International Conference on Machine Learning , year=

  101. [113]

    Notes on the Behavior of MC Dropout , doi =

    Verdoja, Francesco and Kyrki, Ville , year =. Notes on the Behavior of MC Dropout , doi =

  102. [114]

    International Conference on Machine Learning , year=

    Bayesian Learning via Stochastic Gradient Langevin Dynamics , author=. International Conference on Machine Learning , year=

  103. [115]

    International Joint Conference on Artificial Intelligence , year=

    AUC: a Statistically Consistent and more Discriminating Measure than Accuracy , author=. International Joint Conference on Artificial Intelligence , year=

  104. [116]

    Journal of the American Statistical Association , volume =

    Tilmann Gneiting and Adrian E Raftery , title =. Journal of the American Statistical Association , volume =. 2007 , publisher =. doi:10.1198/016214506000001437 , URL =

    2007 doi

  105. [117]

    IEEE transactions on neural networks and learning systems , year=

    A Survey on Deep Active Learning: Recent Advances and New Frontiers , author=. IEEE transactions on neural networks and learning systems , year=

  106. [118]

    International Journal of Computer Vision , volume=

    Semantic understanding of scenes through the ADE20K dataset , author=. International Journal of Computer Vision , volume=. 2019 , publisher=

    2019

  107. [119]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Masked-attention Mask Transformer for Universal Image Segmentation , author=. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2022

  108. [120]

    2021 IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    Swin Transformer: Hierarchical Vision Transformer using Shifted Windows , author=. 2021 IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    2021

  109. [121]

    ArXiv , year=

    FractalNet: Ultra-Deep Neural Networks without Residuals , author=. ArXiv , year=

  110. [122]

    International Conference on Learning Representations , year=

    SGDR: Stochastic Gradient Descent with Warm Restarts , author=. International Conference on Learning Representations , year=

  111. [123]

    LoRA: Low-Rank Adaptation of Large Language Models , author=

  112. [124]

    Packed Ensembles for efficient uncertainty estimation , author=

  113. [125]

    BatchEnsemble: an Alternative Approach to Efficient Ensemble and Lifelong Learning , author=

  114. [126]

    Training independent subnetworks for robust prediction , author=

  115. [127]

    arXiv preprint arXiv:1710.03282 , year=

    Checkpoint ensembles: Ensemble methods from a single training process , author=. arXiv preprint arXiv:1710.03282 , year=

    arXiv

  116. [128]

    International Conference on Learning Representations , year=

    Snapshot Ensembles: Train 1, Get M for Free , author=. International Conference on Learning Representations , year=

  117. [129]

    DeGroot and Stephen E

    Morris H. DeGroot and Stephen E. Fienberg , journal =. The Comparison and Evaluation of Forecasters , urldate =

  118. [130]

    ROC curve, lift chart and calibration plot , volume =

    Vuk, Miha and Curk, Tomaž , year =. ROC curve, lift chart and calibration plot , volume =. Metodološki Zvezki , doi =

  119. [131]

    arXiv preprint arXiv:1412.6980 , year=

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    arXiv

  120. [132]

    International Conference on Learning Representations , year=

    Spectral Normalization for Generative Adversarial Networks , author=. International Conference on Learning Representations , year=

  121. [133]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume =

    Maximum Likelihood from Incomplete Data Via the EM Algorithm , author =. Journal of the Royal Statistical Society: Series B (Methodological) , volume =. 1977 , month =. doi:10.1111/j.2517-6161.1977.tb01600.x , url =

    1977

  122. [134]

    Advances in Neural Information Processing Systems , volume=

    Training certifiably robust neural networks with efficient local lipschitz bounds , author=. Advances in Neural Information Processing Systems , volume=

  123. [135]

    arXiv e-prints , pages=

    Parameter-efficient Bayesian Neural Networks for Uncertainty-aware Depth Estimation , author=. arXiv e-prints , pages=

  124. [136]

    Advances in neural information processing systems , volume=

    Evidential deep learning to quantify classification uncertainty , author=. Advances in neural information processing systems , volume=

  125. [137]

    Mémoires de Mathématique et de Physique, présentés à l’Académie Royale des Sciences par divers savants, Tome sixième , year =

    Pierre-Simon Laplace , title =. Mémoires de Mathématique et de Physique, présentés à l’Académie Royale des Sciences par divers savants, Tome sixième , year =

  126. [138]

    European Conference on Computer Vision , year=

    Uncertainty Estimates and Multi-hypotheses Networks for Optical Flow , author=. European Conference on Computer Vision , year=

  127. [139]

    2021 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Pathobot: A robot for glasshouse crop phenotyping and intervention , author=. 2021 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2021 , organization=

    2021

  128. [140]

    Enhancing decision support in crop production: Analyzing conformal prediction for uncertainty quantification , journal =

    Mohamed Farag and Ahmed Emam and Johannes Leonhardt and Ribana Roscher , keywords =. Enhancing decision support in crop production: Analyzing conformal prediction for uncertainty quantification , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.compag.2025.110559 , url =

    2025

  129. [141]

    Gaussian Stochastic Weight Averaging for Bayesian Low-rank Adaptation of Large Language Models , author=

  130. [142]

    Proceedings of the 41st International Conference on Machine Learning , pages=

    Transitional uncertainty with layered intermediate predictions , author=. Proceedings of the 41st International Conference on Machine Learning , pages=

  131. [143]

    Advances in neural information processing systems , volume=

    Attention is all you need , author=. Advances in neural information processing systems , volume=

  132. [144]

    International Conference on Machine Learning , pages=

    DoRA: Weight-Decomposed Low-Rank Adaptation , author=. International Conference on Machine Learning , pages=. 2024 , organization=

    2024

  133. [145]

    arXiv preprint arXiv:2406.01775 , year=

    Olora: Orthonormal low-rank adaptation of large language models , author=. arXiv preprint arXiv:2406.01775 , year=

    arXiv

  134. [146]

    arXiv preprint arXiv:2310.08659 , year=

    Loftq: Lora-fine-tuning-aware quantization for large language models , author=. arXiv preprint arXiv:2310.08659 , year=

    arXiv

  135. [147]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale , author=

  136. [148]

    Advances in Neural Information Processing Systems , volume=

    Pissa: Principal singular values and singular vectors adaptation of large language models , author=. Advances in Neural Information Processing Systems , volume=

  137. [149]

    One Initialization to Rule them All: Fine-tuning via Explained Variance Adaptation , author=

  138. [150]

    arXiv preprint arXiv:2505.22538 , year=

    Uncertainty Quantification with Proper Scoring Rules: Adjusting Measures to Prediction Tasks , author=. arXiv preprint arXiv:2505.22538 , year=

    arXiv

  139. [151]

    International Conference on Machine Learning , pages=

    On the Practicality of Deterministic Epistemic Uncertainty , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  140. [152]

    Borges, Bruno L C and Pacheco, Bruno M and Silva, Danilo , year =

  141. [153]

    Advances in neural information processing systems , volume=

    SegFormer: Simple and efficient design for semantic segmentation with transformers , author=. Advances in neural information processing systems , volume=

  142. [154]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , year=

    Scene Parsing through ADE20K Dataset , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , year=

  143. [155]

    , author=

    Measuring calibration in deep learning. , author=

  144. [156]

    arXiv preprint arXiv:2405.14438 , year=

    LoRA-Ensemble: Efficient Uncertainty Modelling for Self-attention Networks , author=. arXiv preprint arXiv:2405.14438 , year=

  145. [157]

    ArXiv , year=

    Uncertainty quantification in fine-tuned LLMs using LoRA ensembles , author=. ArXiv , year=

  146. [158]

    Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuning , author=. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Paper...

  147. [159]

    ArXiv , year=

    A Tutorial on Principal Component Analysis , author=. ArXiv , year=

  148. [160]

    Advances in neural information processing systems , volume=

    Can you trust your model's uncertainty? evaluating predictive uncertainty under dataset shift , author=. Advances in neural information processing systems , volume=

  149. [161]

    arXiv preprint arXiv:2501.17917 , year=

    Deep ensembles secretly perform empirical bayes , author=. arXiv preprint arXiv:2501.17917 , year=

    arXiv

  150. [162]

    Advances in neural information processing systems , volume=

    Per-pixel classification is not all you need for semantic segmentation , author=. Advances in neural information processing systems , volume=

  151. [163]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Your vit is secretly an image segmentation model , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  152. [164]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Segment anything , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  153. [165]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Vision transformers for dense prediction , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  154. [166]

    ArXiv , year=

    Uncertainty Quantification with Proper Scoring Rules: Adjusting Measures to Prediction Tasks , author=. ArXiv , year=

  155. [167]

    A Theoretical Review of Modern Robust Statistics

    Loh, Po-Ling. A Theoretical Review of Modern Robust Statistics. Annual Review of Statistics and Its Application. 2025. doi:https://doi.org/10.1146/annurev-statistics-112723-034446

    2025 doi

  156. [168]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    On calibrating semantic segmentation models: Analyses and an algorithm , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  157. [169]

    arXiv preprint arXiv:2410.07170 , year=

    Parameter efficient fine-tuning via explained variance adaptation , author=. arXiv preprint arXiv:2410.07170 , year=

  158. [170]

    IEEE Transactions on Neural Networks and Learning Systems , volume=

    A survey on deep active learning: Recent advances and new frontiers , author=. IEEE Transactions on Neural Networks and Learning Systems , volume=. 2024 , publisher=

    2024

  159. [171]

    The Art of Abstention: Selective Prediction and Error Regularization for Natural Language Processing

    Xin, Ji and Tang, Raphael and Yu, Yaoliang and Lin, Jimmy. The Art of Abstention: Selective Prediction and Error Regularization for Natural Language Processing. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International J...

    2021 doi

  160. [172]

    international conference on machine learning , pages=

    Learning algorithms for active learning , author=. international conference on machine learning , pages=. 2017 , organization=

    2017

  161. [173]

    Sodano and F

    M. Sodano and F. Magistri and J. Behley and C. Stachniss , title =

  162. [174]

    arXiv preprint arXiv:2104.14812 , year=

    Segmentmeifyoucan: A benchmark for anomaly segmentation , author=. arXiv preprint arXiv:2104.14812 , year=

    arXiv

  163. [175]

    Towards a Unified View of Parameter-Efficient Transfer Learning , author=

  164. [176]

    International conference on machine learning , pages=

    Attention is not all you need: Pure attention loses rank doubly exponentially with depth , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  165. [177]

    Findings of the Association for Computational Linguistics: ACL 2022 , pages=

    Moefication: Transformer feed-forward layers are mixtures of experts , author=. Findings of the Association for Computational Linguistics: ACL 2022 , pages=

    2022

  166. [178]

    Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning , author=

  167. [179]

    arXiv preprint arXiv:2312.03732 , year=

    A rank stabilization scaling factor for fine-tuning with lora , author=. arXiv preprint arXiv:2312.03732 , year=

    arXiv

  168. [180]

    arXiv preprint arXiv:2105.02723 , year=

    Do you even need attention? a stack of feed-forward layers does surprisingly well on imagenet , author=. arXiv preprint arXiv:2105.02723 , year=

    arXiv

  169. [181]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    You only need less attention at each stage in vision transformers , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  170. [182]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    When shift operation meets vision transformer: An extremely simple alternative to attention mechanism , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  171. [183]

    Skip-Attention: Improving Vision Transformers by Paying Less Attention , author=

  172. [184]

    Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages=

    Transformer feed-forward layers are key-value memories , author=. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages=

    2021

  173. [185]

    LoRA-Ensemble: Efficient Uncertainty Modelling for Self-attention Networks , author=

  174. [186]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Energy considerations of large language model inference and efficiency optimizations , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  175. [187]

    and Awasthi, L

    Attri, I. and Awasthi, L. K. and Sharma, T. P. and Rathee, P. , title =. Ecological Informatics , volume =. 2023 , doi =

    2023

  176. [188]

    2025 , pages =

    Deep learning in agriculture: Recent advances, challenges, and future directions , journal =. 2025 , pages =. doi:10.1016/j.ecoinf.2025.1011042 , url =

    2025

  177. [189]

    2025 , pages =

    Smart plant disease diagnosis using multiple deep learning and MobileViTv2 , journal =. 2025 , pages =. doi:10.1016/j.atech.2025.100319 , url =

    2025

  178. [190]

    Smart Agricultural Technology , volume =

    Menon, Aparna S and Aravinth, J and Sankaran, Rajendran and Kiran, P , title =. Smart Agricultural Technology , volume =. 2025 , doi =

    2025

  179. [191]

    , title =

    Dimitriadis, Tilmann and Gneiting, Tilmann and Jordan, Alexander I. , title =. Proceedings of the National Academy of Sciences , volume =. 2021 , doi =

    2021

  180. [192]

    LoRA Learns Less and Forgets Less , author=. Trans. Mach. Learn. Res. , year=

  181. [193]

    arXiv preprint arXiv:2301.08727 , year=

    Neural architecture search: Insights from 1000 papers , author=. arXiv preprint arXiv:2301.08727 , year=

    arXiv

  182. [194]

    2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    On Calibrating Semantic Segmentation Models: Analyses and An Algorithm , author=. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2023

  183. [195]

    International Conference on Artificial Intelligence and Statistics , pages=

    Understanding the Learning Dynamics of LoRA: A Gradient Flow Perspective on Low-Rank Adaptation in Matrix Factorization , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2025 , organization=

    2025

  184. [196]

    International Conference on Machine Learning , volume=

    LoRA Training Provably Converges to a Low-Rank Global Minimum or It Fails Loudly (But it Probably Won’t Fail) , author=. International Conference on Machine Learning , volume=

  185. [197]

    Journal of Machine Learning Research , year =

    Nitish Srivastava and Geoffrey Hinton and Alex Krizhevsky and Ilya Sutskever and Ruslan Salakhutdinov , title =. Journal of Machine Learning Research , year =

  186. [198]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Cutmix: Regularization strategy to train strong classifiers with localizable features , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  187. [199]

    International Conference on Machine Learning , pages=

    Spectral normalisation for deep reinforcement learning: an optimisation perspective , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  188. [200]

    arXiv preprint arXiv:2507.13338 , year=

    Training transformers with enforced lipschitz constants , author=. arXiv preprint arXiv:2507.13338 , year=

    arXiv

  189. [201]

    arXiv preprint arXiv:1610.02136 , year=

    A baseline for detecting misclassified and out-of-distribution examples in neural networks , author=. arXiv preprint arXiv:1610.02136 , year=

    arXiv

  190. [202]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    A survey on deep neural network pruning: Taxonomy, comparison, analysis, and recommendations , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2024 , publisher=

    2024

  191. [203]

    Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    Loraprune: Structured pruning meets low-rank parameter-efficient fine-tuning , author=. Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    2024

  192. [204]

    Low-power computer vision , pages=

    A survey of quantization methods for efficient neural network inference , author=. Low-power computer vision , pages=. 2022 , publisher=

    2022

  193. [205]

    Advances in neural information processing systems , volume=

    Qlora: Efficient finetuning of quantized llms , author=. Advances in neural information processing systems , volume=

  194. [206]

    A. P. Dempster and N. M. Laird and D. B. Rubin , journal =. Maximum Likelihood from Incomplete Data via the EM Algorithm , urldate =

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.