Pith. sign in

REVIEW 5 major objections 6 minor 44 references

Few-Shot Optimization for Sensor Data Using Large Language Models: A Case Study on Fatigue Detection

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read For accelerometer fatigue detection, a hybrid selector that filters by Euclidean distance then re-ranks by LLM relevance reaches 69.13% mean macro F1, a relative improvement of 16.6% over random example selection.

desk verdict HED-LM is a clear application of distance-filtering plus LLM re-ranking for sensor data, but the evaluation leaks per-user labels and the paper's own stats contradict whether the 2.3% gain over distance-only is real. read the letter →

arxiv 2505.18754 v1 pith:GPUN3ELT submitted 2025-05-24 cs.CL cs.AIcs.LG

Elsen Ronando , Sozo Inoue This is my paper
classification cs.CLcs.AIcs.LG
keywords Few-shotpromptingExampleselectionLLMscoringFatiguedetectionAccelerometerdataEuclideandistanceHED-LMSensorclassification
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

This paper proposes HED-LM, a two-stage example selector for few-shot prompting of large language models on sensor data. The first stage keeps the labeled samples closest to the test instance by Euclidean distance in a 30-dimensional feature space; the second stage has an LLM score each candidate for numeric similarity and label alignment, re-ranks them, and hands the top two to a 2-shot prompt. The paper argues that this pairing addresses the weakness of each ingredient alone: distance ignores whether a numerically close sample carries the right label, while LLMs handle contextual reasoning but read raw numeric signals poorly. On accelerometer data from 19 runners, HED-LM achieves a mean macro F1 score of 69.13%, a relative improvement of 16.6% over random example selection and 2.3% over distance-only filtering. The intended conclusion is that numeric and contextual similarity are complementary criteria for choosing few-shot examples.

What carries the argument

The load-bearing mechanism is the HED-LM selection pipeline. Each raw 180-sample acceleration trace is windowed into three 60-sample segments, low-pass filtered, and reduced to 30 features (ten per segment, including mean, RMS, skewness, kurtosis, dominant frequency, and low-band energy). A test instance's feature vector is compared against all labeled samples by Euclidean distance; the distance-K closest candidates are then scored by an LLM on a 0-to-1 relevance scale that combines numeric feature differences with label synergy, and the top-K are re-ranked for prompt construction. The prompt itself is 2-shot, carrying one fatigue and one non-fatigue example, plus domain-knowledge threshold rules (for example, 'RMS values above 0.5 in segments 2 and 3 typically indicate fatigue') that the LLM uses during scoring and final classification. The design is what makes the argument run: the distance stage supplies numeric proximity cheaply, and the LLM stage supplies the contextual judgment the distance metric lacks.

What would settle it

Re-run the HED-LM comparison under a strict inductive split: choose two fixed support examples per user from a separate training portion (one fatigue, one non-fatigue) with no access to test-instance labels, and compare HED-LM, distance-only, and random selection. If the reported 16.6% relative gain over random shrinks or reverses under that split, the same-user label access is carrying the result.

Watch

Extended reading notes

Core claim

The paper's central claim, stated in its own terms, is that combining a Euclidean-distance prefilter with LLM-based contextual relevance scoring improves few-shot prompting on high-variability sensor data. On a dataset of 6006 accelerometer strides from 19 runners labeled fatigue or non-fatigue, HED-LM with distance-K=5 and top-K=3 reaches a mean macro F1 of 69.13±10.71%, compared with 59.30±10.13% for random selection and 67.61±11.39% for distance-only filtering. The comparison is run per user, with each test instance's 2-shot prompt built from the same user's other labeled samples, and the paper reports that HED-LM forms a statistically distinct, higher-performing cluster than random selection and a Random Forest baseline, while its post-hoc tests place distance-only and HED-LM in the same performance cluster. The intended moral is that the hybrid selector is robust across users and that domain-knowledge thresholds embedded in the prompts sharpen the LLM's relevance scores from mid-range to decisive values.

Load-bearing premise

The load-bearing premise is that selecting each test instance's two support examples from the same user's other labeled samples is a legitimate few-shot protocol; if that transductive access to nearly all per-user labels is treated as leakage rather than few-shot learning, the headline comparisons against random and distance-only selection would not transfer to a strict train/test split.

Editorial extensions

If this is right

  • Under the paper's protocol, random example selection and a traditional Random Forest trained on two samples form a lower-performing cluster, so few-shot prompting on sensor data is not automatically better than a simple baseline.
  • Most of the hybrid gain is carried by the distance prefilter: the paper's mean improvement over distance-only filtering is 2.3% relative, and its own significance tests put distance-only and HED-LM in the same performance cluster.
  • Two well-chosen examples can beat a full-shot prompt: the paper reports 2-shot HED-LM at 67.70% macro F1 on User 4 versus 52.68% for its full-shot variant, with roughly 7.7 times lower inference time.
  • Domain-knowledge thresholds are doing real work: when they are removed, LLM relevance scores cluster in the 0.4–0.6 range and re-ranking loses power, so the reported gains depend on injecting those rules into the prompts.
  • Per-user evaluation matters for sensor few-shot tasks: macro F1 across individual users in the paper ranges from about 52% to 90% depending on user, so pooled numbers alone hide which subjects benefit from selection.

Reading between the lines

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

  • My reading of the paper's own statistics is that distance-only filtering deserves most of the credit: the paper's post-hoc tests place distance-only and HED-LM in the same high-performance cluster, so a reader should expect the LLM re-ranking to add a small, user-dependent margin rather than a step change.
  • The evaluation is transductive: support examples for each test instance are drawn from the same user's labeled pool, so the method is really selecting which labeled neighbors accompany a query, not learning a two-example training set that transfers to new users; a strict inductive few-shot split would likely narrow the gap over random selection.
  • A cheap falsifying experiment follows from that transductive setting: compare HED-LM against a plain k-nearest-neighbor classifier on the same 30 features; if k-NN matches or beats HED-LM, the LLM scoring stage is not adding information beyond the distance metric.
  • The hand-written domain-knowledge thresholds are a likely source of user-specific sensitivity (the paper's own User-10 failure), so a natural extension is to learn those thresholds from a held-out slice per user and test whether gray-zone cases become less frequent.
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

5 major / 6 minor

Summary. The paper proposes HED-LM, a two-stage example-selection method for few-shot prompting on sensor data: candidates are first filtered by Euclidean distance in a 30-dimensional feature space, then re-ranked by an LLM relevance score that combines numeric similarity with label synergy, and finally a 2-shot prompt is used to classify accelerometer fatigue vs. non-fatigue. Experiments on a 19-user running dataset report mean macro F1 of 69.13 for HED-LM #ParamA against 59.30 for random selection and 67.61 for distance-only selection, which the authors describe as relative improvements of 16.6% and 2.3%. The paper also studies the influence of LLM-generated domain knowledge and reports Friedman/Nemenyi and t/Wilcoxon significance tests.

Significance. The idea of combining a cheap numerical similarity filter with an LLM-based contextual re-ranker for in-context example selection is reasonable and potentially useful for sensor-data classification. The manuscript is transparent in several respects: it provides full prompt templates in the appendix, per-user results, confusion matrices, and a statistical analysis. However, the central claim is not supported as stated. The evaluation protocol is not a few-shot protocol, the paper's own Nemenyi test contradicts its pairwise significance claim for the main 2.3% margin, the domain knowledge is generated from the same dataset's feature values and then applied back to that dataset, and the HED-LM prompt is enriched with LLM score/reason fields that the baselines may not receive. These are load-bearing issues, not presentation fixes.

major comments (5)
  1. [§4.1, Evaluation Protocol and Prompt Construction] The evaluation is not few-shot. The protocol states that for each test instance, 'exactly two support examples drawn from the same user’s data, strictly excluding the test instance itself' are used, meaning the candidate pool for selection is the entire remaining labeled set of that user (e.g., 327 samples for user10). The method therefore has access to the labels of nearly the whole per-user set at inference time, and the LLM re-ranking in Eq. (7) explicitly uses label synergy to choose support examples. This does not demonstrate few-shot generalization from a small, fixed labeled set; it demonstrates per-test-instance retrieval from a large labeled pool. The headline claim of 'few-shot optimization' and the comparisons framed as 'few-shot' are consequently invalidated. The authors should either adopt a genuine few-shot protocol with a fixed support set drawn from a disjoint training split, or reframe the contribution as large-pool example retrieval and remove the few-shot claims.
  2. [§5.1, Tables 4 and 5] The paper contains a direct statistical contradiction. Table 4 reports a Nemenyi post-hoc p-value of 0.5369701 for Distance vs. HED-LM #ParamA, which is not significant, and the text explicitly states that Distance, HED-LM #ParamA, and HED-LM #ParamB 'are potentially in roughly equal performance clusters.' Yet Table 5 reports a t-test p-value of 0.02716 and a Wilcoxon p-value of 0.00240 for the same comparison, and the text claims that HED-LM #ParamA is 'significantly different (and likely superior) to Distance.' These two analyses cannot both be correct as presented. The Nemenyi test is the more conservative, multiple-comparison-corrected procedure, and on its basis the 2.3% mean improvement over distance-only is not statistically significant. The authors must resolve this inconsistency and revise the abstract and Section 1 claims accordingly, since the 2.3% relative improvement is a central quantitative contribution.
  3. [Appendix A.2 and §3.4] The domain knowledge used in LLM scoring and in the few-shot prompts is generated by GPT-4o from the dataset's feature values (Figure A2: 'Based on the attached file, please analyze carefully to create a domain of knowledge'), and the resulting thresholds (e.g., 'Mean Acceleration: Values above 0.44 are indicative of fatigue') are then applied to score and classify the same dataset. This is circular: the scoring rules are fitted to the evaluation data, including its label-related feature distributions, so improvements attributed to 'domain knowledge' may simply reflect test-set information. To support the claim that domain knowledge improves performance, the thresholds must be derived from an independent training split or from external physiological sources, and the test subjects' feature values must not be used to generate the knowledge.
  4. [§3.6 and Appendix A.3] The HED-LM few-shot prompt includes, for each selected example, a 'Relevance Score and Reason' produced by the LLM scoring stage for that test instance. The Random and Distance baselines, which perform no LLM scoring, cannot include such fields, and the paper does not state that they receive equivalent score/reason information. The comparison is therefore confounded: part of HED-LM's gain may come from the additional prompt content (self-referential relevance scores and justifications) rather than from the example-selection mechanism itself. The authors should either provide the baselines with comparable scored examples (e.g., random or distance-based scores with reasons) or run an ablation of HED-LM that removes the score/reason fields from the prompt.
  5. [§4.2.1 and Table 3] The hyperparameters distance-K and top-K are tuned on User ID 4 (Figure 4), and the same user's result in Table 3 (67.70% for #ParamA) is then reported as part of the aggregate mean. This introduces selection bias for that user and makes the mean over the 19 users not a clean evaluation of a fixed configuration. The parameters should be chosen on a validation set disjoint from the test users, or nested cross-validation should be used, with the selection procedure stated explicitly.
minor comments (6)
  1. [§3.1] The text says the dataset 'consists of 6006 labeled subjects'; this should be '6006 labeled instances' or 'samples,' since the dataset has 19 participants.
  2. [Abbreviations] The abbreviation list defines HED-LM as 'Hybrid Edit Distance - Language Model,' but the paper consistently uses 'Hybrid Euclidean Distance with Large Language Models'; the abbreviation definition should match the body.
  3. [§5.1] There is a typo: 'melt detection' should be 'fatigue detection'.
  4. [§4.1] The sentence 'Using a consistent n = 2 across both ML and LLM paradigms ensures a balanced and realistic few-shot comparison' is misleading, because the ML baseline trains on exactly 2 samples while the LLM methods select support examples from a per-user pool of hundreds of labeled samples; the two paradigms do not use the same amount of label information.
  5. [§5.1, Table 4] The Friedman test is reported as 'F-statistic = 54.55'; the Friedman test typically yields a chi-square statistic, and the Iman-Davenport F statistic is a different quantity. Please specify which test was used and report the appropriate statistic with degrees of freedom.
  6. [Conclusion] The conclusion repeats the claim that 'Post-hoc Nemenyi showed that the HED-LM approach (both #ParamA and #ParamB) formed a superior group significantly different from Random and traditional ML methods,' while the body of Section 5.1 also states that Distance, HED-LM #ParamA, and HED-LM #ParamB are not significantly different from each other; the conclusion should be consistent with the corrected post-hoc reading.

Circularity Check

1 steps flagged · score 6.0 of 10

HED-LM's domain knowledge is generated by GPT-4o from the same dataset it classifies, so part of the reported improvement over distance-only is a fitted input rather than an independent prediction; the per-user support selection also makes the protocol transductive.

  1. fitted input called prediction [Appendix A.2, Figs. A2-A3; injected via Fig. A1 and Fig. A4; evaluated in Section 4.2.4]
    "we employ the GPT-4o model in the feature analysis procedure to extract domain knowledge that closely corresponds to the context of the feature values in the sensor data... Based on the attached file, please analyze carefully to create a domain of knowledge... The classification of fatigue and non-fatigue is determined by evaluating specific features across three segments... The key features and their thresholds are described below"

    The thresholds used in LLM scoring and few-shot prompting are not external expert knowledge; they are produced by GPT-4o from the same dataset's feature values, with class-specific numeric criteria such as 'Mean Acceleration: Values above 0.44 are indicative of fatigue.' These thresholds are then injected into the scoring prompt (Fig. A1) and the final few-shot prompt (Fig. A4) used to classify that same dataset. The classifier's 'domain knowledge' is therefore fitted to the evaluation data, so the reported gain of HED-LM over distance-only (Tables 3-5) is partly an artifact of this fitted input. Section 4.2.4 confirms domain knowledge is the operative lever, improving HED-LM #ParamA by +8.08% on User-ID 4.

full rationale

The central derivation of HED-LM is not otherwise circular: Euclidean filtering, LLM re-ranking, and 2-shot prompting are defined independently of the outcome metric, and the baselines use the same protocol. However, the paper's own Appendix A.2 shows that the domain knowledge that drives both the re-ranking (Eq. 7) and the final prompt was generated by GPT-4o from the dataset's feature values, so the 'knowledge' is fitted to the data being classified. Section 4.2.4 demonstrates that this fitted knowledge is responsible for a large share of the improvement (e.g., +8.08% for #ParamA on User-ID 4), meaning the headline 2.3% over distance-only is at least partially circular. Additional non-circularity concerns reinforce the need for caution: Section 4.1 draws the two support examples for each test instance from all other labeled samples of the same user, making the evaluation transductive rather than few-shot; and the statistical support is internally contradictory (Table 4 Nemenyi p=0.537 for Distance vs #ParamA, vs Table 5 t-test p=0.027). Self-citation [11] is used for motivation and model choice but is not the load-bearing reduction here. Score 6 reflects that one central 'prediction' component is a fitted input.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim rests on empirical choices that are fitted to the evaluation data: distance-K/top-K are tuned on User ID 4, the 2-shot setting is chosen from exploratory experiments, and the domain-knowledge thresholds are generated from the same dataset and then reused in prompts. The transductive support-set design is a domain assumption about what counts as few-shot. No new physical entities are introduced.

free parameters (6)
  • distance-K and top-K (#ParamA: Kd=5, Kt=3; #ParamB: Kd=10, Kt=5) = 5/3 and 10/5
    Chosen empirically from a sensitivity analysis on User ID 4; they control how many candidates are retrieved by Euclidean distance and how many survive LLM re-ranking, and directly affect macro F1 and runtime.
  • number of shots in prompt = 2
    Selected based on exploratory experiments comparing 2-shot and full-shot; authors report gains beyond two examples are marginal.
  • low-pass filter cutoff and order = 30 Hz, order 4
    Hand-chosen preprocessing values assumed to preserve fatigue-relevant motion frequencies while removing noise.
  • window segmentation = 3 equal windows of 60 samples
    Fixed segmentation chosen to keep features compact; authors note physiological transitions may not align with rigid boundaries.
  • domain-knowledge thresholds = e.g., RMS above 0.5 in segments 2-3, mean acceleration below 0.31 suggest fatigue; GPT-4o-generated feature thresholds…
    Generated by GPT-4o from the dataset feature values and injected into LLM scoring and few-shot prompts; effectively fitted to the evaluation data.
  • LLM temperature = 0.3
    Model configuration chosen for consistency; not fitted to the target result but affects output variability.
assumptions (6)
  • domain assumption The ground-truth fatigue labels in the Kathirgamanathan dataset are correct and the acceleration magnitude is sufficient for fatigue classification.
    Section 3.1 uses the dataset labels as ground truth and represents each instance by |a|, without external validation of fatigue state.
  • domain assumption Euclidean distance in the 30-dimensional feature space is a valid measure of example relevance for few-shot prompting on sensor data.
    Section 3.3 computes d(x_new, y) and assumes smaller distance implies likely same fatigue label.
  • domain assumption GPT-4o-mini can reliably score numeric similarity and label synergy from tabular feature summaries and follow the scoring rubric.
    Section 3.4 relies on the LLM's contextual reasoning to assign relevance scores; no calibration or validation of the scores is provided.
  • ad hoc to paper Domain knowledge generated by GPT-4o from the dataset features is accurate and generalizes to all subjects.
    Appendix A.2 derives thresholds from feature values; the paper then uses these thresholds in scoring and prediction, but they are not validated on independent data.
  • ad hoc to paper Per-user evaluation with the rest of the user's labeled instances as candidate pool is a valid few-shot protocol.
    Section 4.1: support examples are selected from all other labeled instances of the same user, which is a transductive setting rather than a fixed small labeled support set.
  • standard math Standard statistical machinery (Friedman, Nemenyi, Cliff's delta) is applied correctly to the 19-subject macro F1 scores.
    Used in Section 5.1 to infer method differences; internal inconsistency between Table 4 and Table 5 casts doubt on the analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Few-Shot Optimization for Sensor Data Using Large Language Models: A Case Study on Fatigue Detection." pith.science (2026). https://pith.science/paper/GPUN3ELT

@misc{pith2026250518754,
  author       = {Pith},
  title        = {Pith review of: Few-Shot Optimization for Sensor Data Using Large Language Models: A Case Study on Fatigue Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GPUN3ELT}},
  note         = {Machine review of arXiv:2505.18754}
}
abstract

In this paper, we propose a novel few-shot optimization with HED-LM (Hybrid Euclidean Distance with Large Language Models) to improve example selection for sensor-based classification tasks. While few-shot prompting enables efficient inference with limited labeled data, its performance largely depends on the quality of selected examples. HED-LM addresses this challenge through a hybrid selection pipeline that filters candidate examples based on Euclidean distance and re-ranks them using contextual relevance scored by large language models (LLMs). To validate its effectiveness, we apply HED-LM to a fatigue detection task using accelerometer data characterized by overlapping patterns and high inter-subject variability. Unlike simpler tasks such as activity recognition, fatigue detection demands more nuanced example selection due to subtle differences in physiological signals. Our experiments show that HED-LM achieves a mean macro F1-score of 69.13$\pm$10.71%, outperforming both random selection (59.30$\pm$10.13%) and distance-only filtering (67.61$\pm$11.39%). These represent relative improvements of 16.6% and 2.3%, respectively. The results confirm that combining numerical similarity with contextual relevance improves the robustness of few-shot prompting. Overall, HED-LM offers a practical solution to improve performance in real-world sensor-based learning tasks and shows potential for broader applications in healthcare monitoring, human activity recognition, and industrial safety scenarios.

Figures

Figures reproduced from arXiv: 2505.18754 by the authors.

Figure 1
Figure 1. Comparison of fatigue and non-fatigue accelerometer signals from three different users [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Proposed HED-LM framework for physical fatigue detection [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Comparison of evaluation pipelines between the ML baseline and LLM-based approaches. 4.2. Experimental Results 4.2.1. Impact of Distance-K and Top-K Parameters Before analyzing the final performance comparison with baseline methods, we first evaluated the effect of key parameters on the performance of HED-LM. In particular, we compared four config￾urations: #ParamA with distanceK = 5, topK = 3, #ParamB with distance… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of HED-LM performance and computation time across different parameter configu [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: Macro F1-score comparison for different methods using 2-shot and full-shot setups on User ID 4 [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: Computation time comparison between 2-shot and full-shot configurations for different methods on [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: Macro F1-score performance comparison: Influence of domain knowledge on User-ID 4 [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: Overall user comparison: HED-LM and its parameters versus baselines [PITH_FULL_IMAGE:figures/full_fig_p023_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 23 canonical work pages

  1. [1]

    Language Models are Few-Shot Learners, 2020, [arXiv:cs.CL/2005.14165]

    Brown, T.B.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J.; Dhariwal, P .; Neelakantan, A.; Shyam, P .; Sastry, G.; Askell, A.; et al. Language Models are Few-Shot Learners, 2020, [arXiv:cs.CL/2005.14165]

    arXiv 2020

  2. [2]

    Atlas: Few-shot Learning with Retrieval Augmented Language Models, 2022, [arXiv:cs.CL/2208.03299]

    Izacard, G.; Lewis, P .; Lomeli, M.; Hosseini, L.; Petroni, F.; Schick, T.; Dwivedi-Yu, J.; Joulin, A.; Riedel, S.; Grave, E. Atlas: Few-shot Learning with Retrieval Augmented Language Models, 2022, [arXiv:cs.CL/2208.03299]

    arXiv 2022

  3. [3]

    Few-shot Learning with Multilingual Generative Language Models

    Lin, X.V .; Mihaylov, T.; Artetxe, M.; Wang, T.; Chen, S.; Simig, D.; Ott, M.; Goyal, N.; Bhosale, S.; Du, J.; et al. Few-shot Learning with Multilingual Generative Language Models. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2021

    work page 2021

  4. [4]

    Concept Learners for Few-Shot Learning, 2021, [arXiv:cs.LG/2007.07375]

    Cao, K.; Brbic, M.; Leskovec, J. Concept Learners for Few-Shot Learning, 2021, [arXiv:cs.LG/2007.07375]

    arXiv 2021

  5. [5]

    Bi-Level Meta-Learning for Few-Shot Domain Generalization

    Qin, X.; Song, X.; Jiang, S. Bi-Level Meta-Learning for Few-Shot Domain Generalization. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2023, pp. 15900–15910. https://doi.org/10.110 9/CVPR52729.2023.01526

    arXiv 2023

  6. [6]

    Making Pre-trained Language Models Better Few-shot Learners, 2021, [arXiv:cs.CL/2012.15723]

    Gao, T.; Fisch, A.; Chen, D. Making Pre-trained Language Models Better Few-shot Learners, 2021, [arXiv:cs.CL/2012.15723]

    arXiv 2021

  7. [7]

    Quantifying Language Models’ Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting

    Sclar, M.; Choi, Y.; Tsvetkov, Y.; Suhr, A. Quantifying Language Models’ Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting. In Proceedings of the The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  8. [8]

    Designing Informative Metrics for Few-Shot Example Selection

    Adiga, R.; Subramanian, L.; Chandrasekaran, V . Designing Informative Metrics for Few-Shot Example Selection. In Proceedings of the Findings of the Association for Computational Linguistics: ACL 2024; Ku, 42 of 43 L.W.; Martins, A.; Srikumar, V ., Eds., Bangkok, Thailand, 2024; pp. 10127–10135. https://doi.org/10.18653 /v1/2024.findings-acl.602

    work page 2024

Show all 44 references
  1. [9]

    Active Instance Selection for Few-Shot Classification

    Shin, J.; Kang, Y.; Jung, S.; Choi, J. Active Instance Selection for Few-Shot Classification. IEEE Access 2022, 10, 133186–133195. https://doi.org/10.1109/ACCESS.2022.3231365

    2022

  2. [10]

    Few-Shot Learning-Based Human Activity Recognition, 2019, [arXiv:cs.LG/1903.10416]

    Feng, S.; Duarte, M.F. Few-Shot Learning-Based Human Activity Recognition, 2019, [arXiv:cs.LG/1903.10416]

    2019 arXiv

  3. [11]

    Leveraging Large Language Models to Enhance Understanding of Accelerometer Data on Physical Fatigue Detection Question Answering

    Ronando, E.; Inoue, S. Leveraging Large Language Models to Enhance Understanding of Accelerometer Data on Physical Fatigue Detection Question Answering. In Proceedings of the 112th Mobile Computing and New Social Systems, 83rd Ubiquitous Computing Systems, 41st Consumer Device...

    2024

  4. [12]

    CTYUN-AI at SemEval-2024 Task 7: Boosting Numerical Understanding with Limited Data Through Effective Data Alignment

    Fan, Y.; Yang, D.; He, X. CTYUN-AI at SemEval-2024 Task 7: Boosting Numerical Understanding with Limited Data Through Effective Data Alignment. In Proceedings of the Proceedings of the 18th International Workshop on Semantic Evaluation (SemEval-2024); Ojha, A.K.; Do˘ gruöz, A....

    2024

  5. [13]

    The first step is the hardest: Pitfalls of Representing and Tokenizing Temporal Data for Large Language Models

    Spathis, D.; Kawsar, F. The first step is the hardest: Pitfalls of Representing and Tokenizing Temporal Data for Large Language Models. Journal of the American Medical Informatics Association : JAMIA 2023. https://doi.org/10.48550/arXiv.2309.06236

    2023 doi arXiv

  6. [14]

    Evaluating Large Language Models as Virtual Annotators for Time-series Physical Sensing Data

    Hota, A.; Chatterjee, S.; Chakraborty, S. Evaluating Large Language Models as Virtual Annotators for Time-series Physical Sensing Data. ACM Transactions on Intelligent Systems and Technology 2024. https: //doi.org/10.1145/3696461

    2024 doi

  7. [15]

    SensorLLM: Aligning Large Language Models with Motion Sensors for Human Activity Recognition, 2024, [arXiv:cs.CL/2410.10624]

    Li, Z.; Deldari, S.; Chen, L.; Xue, H.; Salim, F.D. SensorLLM: Aligning Large Language Models with Motion Sensors for Human Activity Recognition, 2024, [arXiv:cs.CL/2410.10624]

    2024 arXiv

  8. [16]

    Small Data, Big Challenges: Pitfalls and Strategies for Machine Learning in Fatigue Detection

    Jeworutzki, A.; Schwarzer, J.; Von Luck, K.; Stelldinger, P .; Draheim, S.; Wang, Q. Small Data, Big Challenges: Pitfalls and Strategies for Machine Learning in Fatigue Detection. In Proceedings of the Proceedings of the 16th International Conference on PErvasive Technologies ...

    2023

  9. [17]

    Calibrate Before Use: Improving Few-Shot Performance of Language Models, 2021, [arXiv:cs.CL/2102.09690]

    Zhao, T.Z.; Wallace, E.; Feng, S.; Klein, D.; Singh, S. Calibrate Before Use: Improving Few-Shot Performance of Language Models, 2021, [arXiv:cs.CL/2102.09690]

    2021 arXiv

  10. [18]

    Sensorimotor distance: A grounded measure of semantic similarity for 800 million concept pairs

    Cai, W.; Louise, C. Sensorimotor distance: A grounded measure of semantic similarity for 800 million concept pairs. Behavior Research Methods 2022, 55, 3416–3432. https://doi.org/10.3758/s13428-022-01965-7

    2022 doi

  11. [19]

    Knowledge Prompting for Few-shot Action Recognition, 2022, [arXiv:cs.CV/2211.12030]

    Shi, Y.; Wu, X.; Lin, H. Knowledge Prompting for Few-shot Action Recognition, 2022, [arXiv:cs.CV/2211.12030]

    2022 arXiv

  12. [20]

    A Good Prompt Is Worth Millions of Parameters: Low-resource Prompt-based Learning for Vision-Language Models, 2022, [arXiv:cs.CV/2110.08484]

    Jin, W.; Cheng, Y.; Shen, Y.; Chen, W.; Ren, X. A Good Prompt Is Worth Millions of Parameters: Low-resource Prompt-based Learning for Vision-Language Models, 2022, [arXiv:cs.CV/2110.08484]

    2022 arXiv

  13. [21]

    Noisy Channel Language Model Prompting for Few-Shot Text Classification, 2022, [arXiv:cs.CL/2108.04106]

    Min, S.; Lewis, M.; Hajishirzi, H.; Zettlemoyer, L. Noisy Channel Language Model Prompting for Few-Shot Text Classification, 2022, [arXiv:cs.CL/2108.04106]

    2022 arXiv

  14. [22]

    Selecting Shots for Demographic Fairness in Few-Shot Learning with Large Language Models, 2023, [arXiv:cs.CL/2311.08472]

    Aguirre, C.; Sasse, K.; Cachola, I.; Dredze, M. Selecting Shots for Demographic Fairness in Few-Shot Learning with Large Language Models, 2023, [arXiv:cs.CL/2311.08472]

    2023 arXiv

  15. [23]

    Use Random Selection for Now: Investigation of Few-Shot Selection Strategies in LLM-based Text Augmentation for Classification, 2024, [arXiv:cs.CL/2410.10756]

    Cegin, J.; Pecher, B.; Simko, J.; Srba, I.; Bielikova, M.; Brusilovsky, P . Use Random Selection for Now: Investigation of Few-Shot Selection Strategies in LLM-based Text Augmentation for Classification, 2024, [arXiv:cs.CL/2410.10756]

    2024 arXiv

  16. [24]

    True Few-Shot Learning with Language Models

    Perez, E.; Kiela, D.; Cho, K. True Few-Shot Learning with Language Models. ArXiv 2021, abs/2105.11447

    2021 arXiv

  17. [25]

    On Training Instance Selection for Few-Shot Neural Text Generation

    Chang, E.; Shen, X.; Yeh, H.S.; Demberg, V . On Training Instance Selection for Few-Shot Neural Text Generation. ArXiv 2021, abs/2107.03176

    2021 arXiv

  18. [26]

    Active Learning Principles for In-Context Learning with Large Language Models, 2023, [arXiv:cs.CL/2305.14264]

    Margatina, K.; Schick, T.; Aletras, N.; Dwivedi-Yu, J. Active Learning Principles for In-Context Learning with Large Language Models, 2023, [arXiv:cs.CL/2305.14264]

    2023 arXiv

  19. [27]

    More Samples or More Prompts? Exploring Effective In-Context Sampling for LLM Few-Shot Prompt Engineering, 2024, [arXiv:cs.CL/2311.09782]

    Yao, B.; Chen, G.; Zou, R.; Lu, Y.; Li, J.; Zhang, S.; Sang, Y.; Liu, S.; Hendler, J.; Wang, D. More Samples or More Prompts? Exploring Effective In-Context Sampling for LLM Few-Shot Prompt Engineering, 2024, [arXiv:cs.CL/2311.09782]

    2024 arXiv

  20. [28]

    Automatic Combination of Sample Selection Strategies for Few-Shot Learning, 2024, [arXiv:cs.LG/2402.03038]

    Pecher, B.; Srba, I.; Bielikova, M.; Vanschoren, J. Automatic Combination of Sample Selection Strategies for Few-Shot Learning, 2024, [arXiv:cs.LG/2402.03038]

    2024 arXiv

  21. [29]

    In-Context Learning with Iterative Demonstration Selection

    Qin, C.; Zhang, A.; Chen, C.; Dagar, A.; Ye, W. In-Context Learning with Iterative Demonstration Selection. In Proceedings of the Findings of the Association for Computational Linguistics: EMNLP 2024; Al-Onaizan, Y.; Bansal, M.; Chen, Y.N., Eds., Miami, Florida, USA, 2024; pp....

    2024

  22. [30]

    Skill-Based Few-Shot Selection for In-Context Learning

    An, S.; Zhou, B.; Lin, Z.; Fu, Q.; Chen, B.; Zheng, N.; Chen, W.; Lou, J.G. Skill-Based Few-Shot Selection for In-Context Learning. In Proceedings of the Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing; Bouamor, H.; Pino, J.; Bali, K., Ed...

    2023 doi

  23. [31]

    Large Language Models are Few-Shot Health Learners, 2023, [arXiv:cs.CL/2305.15525]

    Liu, X.; McDuff, D.; Kovacs, G.; Galatzer-Levy, I.; Sunshine, J.; Zhan, J.; Poh, M.Z.; Liao, S.; Achille, P .D.; Patel, S. Large Language Models are Few-Shot Health Learners, 2023, [arXiv:cs.CL/2305.15525]

    2023 arXiv

  24. [32]

    A Few-Shot Learning Based Fault Diagnosis Model Using Sensors Data from Industrial Machineries

    Siraj, F.M.; Ayon, S.T.K.; Uddin, J. A Few-Shot Learning Based Fault Diagnosis Model Using Sensors Data from Industrial Machineries. Vibration 2023, 6, 1004–1029. https://doi.org/10.3390/vibration6040059

    2023 doi

  25. [33]

    Generating Explanations to understand Fatigue in Runners Using Time Series Data from Wearable Sensors

    Kathirgamanathan, B.; Cunningham, P . Generating Explanations to understand Fatigue in Runners Using Time Series Data from Wearable Sensors. In Proceedings of the ICML 3rd Workshop on Interpretable Machine Learning in Healthcare (IMLH), 2023

    2023

  26. [34]

    Application of Fourier Transform and Butterworth Filter in Signal Denoising

    Zhang, X.; Jiang, S. Application of Fourier Transform and Butterworth Filter in Signal Denoising. In Proceedings of the 2021 6th International Conference on Intelligent Computing and Signal Processing (ICSP), 2021, pp. 1277–1281. https://doi.org/10.1109/ICSP51882.2021.9408933

    2021

  27. [35]

    A Study on the Influence of Sensors in Frequency and Time Domains on Context Recognition.Sensors 2023, 23

    de Souza, P .; Silva, D.; de Andrade, I.; Dias, J.; Lima, J.P .; Teichrieb, V .; Quintino, J.P .; da Silva, F.Q.B.; Santos, A.L.M. A Study on the Influence of Sensors in Frequency and Time Domains on Context Recognition.Sensors 2023, 23. https://doi.org/10.3390/s23125756

    2023 doi

  28. [36]

    A survey on distance and similarity measures for time-series data analysis

    Chen, Z.; Li, H. A survey on distance and similarity measures for time-series data analysis. Information Sciences 2020, 546, 441–465. https://doi.org/10.1016/j.ins.2020.07.070

    2020 doi

  29. [37]

    Selective review of offline change point detection methods

    Truong, C.; Oudre, L.; Vayatis, N. Selective review of offline change point detection methods. Signal Processing 2020, 167, 107299. https://doi.org/10.1016/j.sigpro.2019.107299

    2020

  30. [38]

    Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? In Proceedings of the EMNLP 2022, 2022, pp

    Min, S.; Lewis, M.; Zettlemoyer, L.; Hajishirzi, H. Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? In Proceedings of the EMNLP 2022, 2022, pp. 11044–11064

    2022

  31. [39]

    Adaptive weighting and nearest neighbor-based area control for imbalanced data classification

    Xue, W.; Duan, L.; Hong, X.; Zheng, X. Adaptive weighting and nearest neighbor-based area control for imbalanced data classification. Applied Soft Computing 2025, 177, 113171. https://doi.org/https: //doi.org/10.1016/j.asoc.2025.113171

    2025

  32. [40]

    A Muscle Fatigue Classification Model Based on LSTM and Improved Wavelet Packet Threshold

    Wang, J.; Sun, S.; Sun, Y. A Muscle Fatigue Classification Model Based on LSTM and Improved Wavelet Packet Threshold. Sensors 2021, 21. https://doi.org/10.3390/s21196369

    2021 doi

  33. [41]

    Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting

    Zhou, H.; Zhang, S.; Peng, J.; Zhang, S.; Li, J.; Xiong, H.; Zhang, W. Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting. In Proceedings of the AAAI 2021, 2021, pp. 11106–11115

    2021

  34. [42]

    Meta-Learning in Neural Networks: A Survey.IEEE Transactions on Pattern Analysis and Machine Intelligence 2021

    Hospedales, T.; Antoniou, A.; Micaelli, P .; Storkey, A. Meta-Learning in Neural Networks: A Survey.IEEE Transactions on Pattern Analysis and Machine Intelligence 2021. https://doi.org/10.1109/TPAMI.2021.3079209

    2021

  35. [43]

    The Power of Scale for Parameter-Efficient Prompt Tuning

    Lester, B.; Al-Rfou, R.; Constant, N. The Power of Scale for Parameter-Efficient Prompt Tuning. In Proceedings of the EMNLP 2021, 2021, pp. 3045–3059

    2021

  36. [44]

    Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

    Lewis, P .; Perez, E.; Piktus, A.e.a. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. In Proceedings of the NeurIPS 2020, 2020. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual aut...

    2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.