Pith. sign in

REVIEW 5 major objections 7 minor 28 references

Adaptable Embeddings Network (AEN)

T0 review · 5 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A small dual-encoder classifier matches 10x larger LLMs at 16x lower compute, the paper claims.

desk verdict A genuinely clever KDE-based dual-encoder mechanism that deserves further study, but the paper's central accuracy claim is undermined by its evaluation design. read the letter →

arxiv 2411.13786 v1 pith:4KNIANX7 submitted 2024-11-21 cs.LG cs.CL

Stan Loosmore , Alexander Titus This is my paper · ORCID
classification cs.LGcs.CL
keywords AdaptableEmbeddingsNetworkKernelDensityEstimationdual-encodertextclassificationedgecomputingsyntheticdatazero-shotefficientinference
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 introduces Adaptable Embeddings Networks (AEN), a dual-encoder text classifier that accepts a natural-language condition at runtime and returns a binary verdict on whether a statement satisfies it. The authors claim that, on a synthetic benchmark, AEN's accuracy is comparable to, and in some settings better than, autoregressive models roughly ten times its size, while using about sixteen times fewer FLOPs per input. The design goal is edge deployment: conditions can be preprocessed and cached, so only the statement encoder runs on new text. If the accuracy claim holds on real data, AEN would offer a way to retarget a classifier to new criteria without retraining and without large language model inference costs.

What carries the argument

Kernel Density Estimation applied per embedding dimension is the mechanism that carries the argument. For each of the 384 output dimensions, the statement's attended token embeddings form a separate univariate Gaussian-kernel density estimate (bandwidth by Scott's rule); the condition's mean-pooled embedding is evaluated at each density, producing a vector of log-likelihood-style scores that the classification head consumes. This replaces cosine similarity or concatenated mean-pooled vectors with a distributional comparison, motivated by the paper's finding that per-dimension token distributions are mostly non-parametric.

What would settle it

Run AEN and a fine-tuned or few-shot-tuned LLaMA-3.2-3B on a human-annotated corpus of real conversational statements paired with natural-language conditions, and compare F1; if AEN's margin over the adapted SLM shrinks or reverses, the comparable-or-superior accuracy claim fails. A cheaper check is to have human annotators relabel a sample of the paper's own generated statements and recompute AEN's F1 on the corrected labels.

Watch

Extended reading notes

Core claim

The central claim is that a small, non-autoregressive dual-encoder network can perform zero-shot natural-language classification as accurately as much larger autoregressive models, with dramatically lower compute. AEN encodes the statement with one sentence-transformer and the condition with another; instead of mean-pooling both, it turns every token dimension of the statement into a one-dimensional kernel density estimate and scores the condition's mean-pooled vector against those densities. A small feed-forward head then maps the per-dimension probability scores to a binary classification. The paper reports F1 of 0.74 on its 5,000-sample synthetic holdout, against 0.42–0.54 for quantized LLaMA-3.2-3B and Phi-3.5-mini zero-shot baselines, at roughly 16x lower FLOPs per pass.

Load-bearing premise

The accuracy comparison assumes AEN's synthetic test set is a fair proxy for real-world classification, and that zero-shot use of much larger models is the right baseline; the larger models were not trained or adapted to the task's condition format.

Editorial extensions

If this is right

  • AEN can be used for real-time monitoring where transcription streams are checked against many cached semantic criteria simultaneously.
  • Condition embeddings can be computed once and cached, roughly halving runtime compute and enabling dynamic criterion updates on-device.
  • Because the model is non-autoregressive and tiny relative to SLMs, it fits edge and mobile deployments that cannot run a multi-billion-parameter LLM.
  • The architecture extends to decision trees whose branching conditions are natural-language semantic predicates, though the paper leaves criterion selection to future work.

Reading between the lines

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

  • The headline accuracy comparison is against zero-shot baselines that were never adapted to the synthetic task distribution; a like-for-like comparison would fine-tune or prompt-tune the baselines on the same training data before measuring F1.
  • The labels themselves come from GPT-4o-mini, and the paper concedes human evaluation revealed accuracy gaps; if those label errors are correlated with condition phrasing, AEN's reported F1 could shift on human-annotated data.
  • Since the KDE step is non-parametric and acts on fixed token embeddings, the same head could be retrained on new label distributions with far fewer examples than an LLM fine-tune, a testable extension.
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 / 7 minor

Summary. The paper proposes the Adaptable Embeddings Network (AEN), a dual-encoder classifier for binary text classification against natural-language conditions specified at runtime. A statement encoder produces per-token embeddings while a criterion encoder produces a mean-pooled condition embedding; in the final KDE variant, one encoder's token embeddings are summarized into per-dimension univariate kernel density estimates and the other encoder's mean-pooled vector is scored against those densities, with a small feed-forward head emitting the binary decision. Because condition embeddings can be precomputed and cached, new statements can be classified without retraining and without re-encoding the criterion. All data are synthetic: statements, conditions, and labels are generated with GPT-4o-mini (Section 3.1.2), yielding 2.5M training triples and a 5,000-sample holdout from the same pipeline. On that holdout the paper reports F1 0.74 for AEN versus 0.38-0.54 for zero-shot LLaMA-3.2-3B and Phi-3.5-mini at roughly 16x lower FLOPs per pass, and concludes that AEN matches or exceeds autoregressive models an order of magnitude larger while supporting runtime criterion changes.

Significance. If the empirical claims were reliable, AEN would be a practically interesting contribution: runtime-swappable classification criteria, a genuinely parameter-free KDE feature transform (bandwidth from Scott's rule, no learned density parameters), and concrete, checkable efficiency numbers in Table 11. The adaptability and caching story follows transparently from the architecture and is the strongest part of the paper; the efficiency half of the claim is plausible and falsifiable. I concur with the reader's assessment that the accuracy half is not established: the F1 of 0.74 is selected on the test set (Section 4.1), the baselines are zero-shot on a distribution on which AEN was trained (Section 4.2), and the ground-truth labels are GPT-4o-mini outputs whose accuracy the authors concede is imperfect (Section 6.2). The paper's honest disclosure of labeling gaps is commendable, but it directly undercuts the headline accuracy claim, which needs a redone evaluation before the contribution can be assessed.

major comments (5)
  1. [4.1] Section 4.1 opens by saying that the authors 'examine each hyper parameter of our models and take the highest performing option of each,' and every comparison in that section (Tables 1-9) reports test loss and test F1. Selecting hyperparameters on the same test set that later yields the headline F1 of 0.74 in Table 10 is a form of test-set overfitting; the reported F1 is not an unbiased estimate of AEN's accuracy, and the margin over the zero-shot baselines is correspondingly overstated. The hyperparameter sweeps must be run on a held-out validation split, with the final test set used exactly once.
  2. [4.2, Table 10] The comparison in Table 10 is not like-for-like. AEN is trained on 2.5M statement-condition-label triples drawn from the same synthetic pipeline (Section 3.1) that generates the 5,000-sample evaluation set, while LLaMA-3.2-3B and Phi-3.5-mini are evaluated zero-shot with no exposure to this task distribution. The F1 gap (0.74 vs 0.38-0.54) therefore conflates semantic competence with in-distribution fit. To support the abstract's 'comparable and in certain cases superior' claim, the evaluation should include baselines with comparable in-distribution adaptation (few-shot prompting or light fine-tuning) and, ideally, a real-world or out-of-distribution holdout to test whether the advantage survives outside the synthetic pipeline.
  3. [3.1.2, 6.2] The labels used both for training AEN and for scoring every model were generated by GPT-4o-mini (Section 3.1.2), and Section 6.2 concedes that 'human evaluation revealed accuracy gaps' in those labels. AEN is trained to reproduce this particular labeler's judgments, so the reported F1 measures agreement with that labeler rather than classification accuracy in an absolute sense; the baselines are being scored against the same flawed ground truth. The authors should report AEN's agreement with the human-evaluated subset, compare against an independent labeler if feasible, and state the size of the human-evaluated sample.
  4. [4.1.4, Tables 5-6] Section 4.1.4 states that 'the larger the weight the more precision and less recall the model has initially as shown in 6,' but Table 6 shows the opposite: weight 6 gives precision 0.367-0.426 and recall 0.929-0.945, while weight 1 gives precision 0.685-0.691 and recall 0.475-0.678. In addition, Table 5 shows the chosen weight of 6 (F1 0.584) to be the worst of the three options tested (F1 0.654 and 0.692), which contradicts the paper's own rule of taking the highest-performing option. The loss-weight selection for the final model must be corrected and justified.
  5. [3.2.3, 6.5; Tables 1, 4-8] The paper claims the KDE transform 'proved as effective if not more so' than feeding standard concatenated outputs into a head (Section 6.5), but no matched head-to-head comparison is provided. The numbers that are reported point the other way: the classical feed-forward model reaches F1 0.752 at 2.5M datapoints (Table 1), while the KDE-based AEN variants in Tables 4, 5, and 8 plateau between 0.584 and 0.692. The exact configuration that produces the F1 of 0.74 in Table 10 is never specified as a single hyperparameter set, which also prevents reproduction of the headline result.
minor comments (7)
  1. [4.1.6] The text refers to 'Figure 8' when reporting KDE-function results, but the results appear in Table 8.
  2. [3.1.2] The sentence 'We found that generating them sequentially yielded a class imbalance of 1 to 6 compared a significantly higher proportion then in parallel' is grammatically garbled and leaves unclear whether sequential generation produced or avoided the 1:6 imbalance.
  3. [3.1.2] Condition generation is described as using 'the higher temperature compared to statement generation (1 vs 1.6)', but 1.0 is lower than 1.6; the text should say 'lower temperature'.
  4. [Table 11] The 219M parameter count reported for AEN is not reconciled with the 22.7M (MiniLM) and 109M (mpnet-base) encoder sizes in Table 2; please state how the total is computed.
  5. [6.5] The claim of achieving 'the first natural language adaptable binary classifier' is unsupported and should be removed or substantially qualified in light of existing NLI-style adaptive classifiers.
  6. [Throughout] The manuscript contains numerous typos (e.g., 'discreet' for 'discrete' in Section 1, 'there' for 'their' in Sections 4.1.1-4.1.2, 'exsist' in Section 2.2.2) and never collects the final AEN hyperparameters into a single configuration table; a thorough proofread and a final-configuration table would materially improve reproducibility.
  7. [Data availability] No statement of code or data availability is provided, which is notable because the paper emphasizes the effort invested in the synthetic dataset as a contribution.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: AEN's accuracy is measured, not derived from its training objective, and the evaluation bias (same synthetic pipeline for training and testing, acknowledged label gaps) is a validity concern rather than a definitional reduction.

full rationale

The paper's central derivation is the KDE-based dual-encoder architecture, which is defined independently of the measured outcomes: the KDE transform is parameter-free (bandwidth chosen by Scott's rule, which depends only on token count and standard deviation), and the reported F1 scores are empirically measured on a held-out sample rather than algebraically forced by the training setup. The main weakness is evaluation bias, not circularity. Section 4.2 compares AEN, trained on 2.5 million synthetic statement-condition-label triples, against LLaMA-3.2-3B and Phi-3.5-mini used zero-shot on 5,000 additional samples generated by the same synthetic pipeline, and Section 6.2 concedes that 'human evaluation revealed accuracy gaps' in the GPT-4o-mini labeling. This undermines the external validity of the headline 'comparable or superior' accuracy claim, and Section 4.1's test-set-based hyperparameter selection further inflates reported numbers, but neither issue makes a prediction equal to its input by construction. There are no load-bearing self-citations: the cited related work (SBERT, KDE tutorials, synthetic data surveys) is external and does not smuggle in the AEN mechanism. No circular step can be exhibited with a quote and a specific reduction, so the appropriate circularity score is 0.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the synthetic data distribution and label quality assumptions rather than on a mathematical derivation. The KDE mechanism itself adds no free parameters (bandwidth uses Scott's rule), but every architecture choice around it was selected on test performance, which is a form of fitting.

free parameters (7)
  • KDE bandwidth rule = Scott's rule, chosen over Silverman (F1 0.637 vs 0.607)
    Section 4.1.5; the bandwidth selection method was chosen based on test F1, making it a fitted modeling choice rather than a fixed prior.
  • KDE kernel function = Gaussian, chosen over Epanechnikov and Triangular (F1 0.637 vs 0.611 vs 0.600)
    Section 4.1.6; kernel choice selected on test performance.
  • Which encoder receives KDE transform = Statement encoder, chosen over condition encoder (F1 0.637 vs 0.636)
    Section 4.1.7; decision made from test performance.
  • Learning rate for AEN = 1e-5, chosen over 2e-6 (F1 0.665 vs 0.637)
    Section 4.1.3 Table 4; learning rate selected based on test F1.
  • Loss weight for positive class = 1 (best F1 0.692 in Table 5), though text says 'we use 6 as the loss weight'
    Section 4.1.4; inconsistent reporting and selection based on test F1.
  • Number of training datapoints = 2.5M (best F1 0.752 in Table 1)
    Section 4.1.1; data scale chosen based on test performance.
  • Classification head architecture = Small dense network with batch norm and ReLU, chosen over alternatives by search
    Section 4.1.8 'Head Network Parameters'; not fully specified.
assumptions (5)
  • domain assumption Each dimension of an embedding corresponds to a distinct semantic attribute, so per-dimension KDEs capture meaningful comparison signal.
    Stated as a hypothesis in Section 3.2.3 ('This method was motivated by our hypothesis that individual embedding dimensions represent distinct attributes of the input'); if false, the KDE transform's benefit is unexplained.
  • domain assumption Synthetic statement-condition-label triples generated by GPT-4o-mini form a valid training and evaluation distribution for the claimed text-classification capability.
    Section 3.1.2 and 4.2; all experiments are on this synthetic distribution; Section 6.2 admits real-world alignment is future work.
  • domain assumption GPT-4o-mini labels (temperature 0) are reliable enough to serve as ground truth for F1 measurement.
    Section 3.1.2 'Labeling Generation' uses GPT-4o-mini for labels; Section 6.2 states 'human evaluation revealed accuracy gaps', so label noise is acknowledged.
  • standard math Standard KDE definitions and Scott's rule for bandwidth (h = n^{-1/5} sigma) are valid for per-dimension token embeddings.
    Section 2.3 and 4.1.5; these are standard results used without proof.
  • domain assumption A zero-shot prompt for LLaMA and Phi is the appropriate baseline; fine-tuning these models on the same synthetic data would not change the comparison.
    Section 4.2; no training or few-shot adaptation is performed for the baselines, while AEN is trained on the evaluation distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptable Embeddings Network (AEN)." pith.science (2026). https://pith.science/paper/4KNIANX7

@misc{pith2026241113786,
  author       = {Pith},
  title        = {Pith review of: Adaptable Embeddings Network (AEN)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4KNIANX7}},
  note         = {Machine review of arXiv:2411.13786}
}
read the original abstract

Modern day Language Models see extensive use in text classification, yet this comes at significant computational cost. Compute-effective classification models are needed for low-resource environments, most notably on edge devices. We introduce Adaptable Embeddings Networks (AEN), a novel dual-encoder architecture using Kernel Density Estimation (KDE). This architecture allows for runtime adaptation of classification criteria without retraining and is non-autoregressive. Through thorough synthetic data experimentation, we demonstrate our model outputs comparable and in certain cases superior results to that of autoregressive models an order of magnitude larger than AEN's size. The architecture's ability to preprocess and cache condition embeddings makes it ideal for edge computing applications and real-time monitoring systems.

Figures

Figures reproduced from arXiv: 2411.13786 by the authors.

Figure 1
Figure 1. 1D Kernel Density Estimation showing in [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Data Generation Process Generating synthetic data through LMs presents a significant challenge: ensuring both data diversity and maintaining coherence (Veselovsky et al. [2023]). In our approach, we prioritized diversity of potential transcripts while implementing controlled patterns that allow us to prove convergence on data meeting our defined requirements. These patterns are further detailed below. We leave the e… view at source ↗
Figure 3
Figure 3. AEN General training architecture 3.2.1 Data Preparation Our data preparation process involves two main steps: Batching Strategy To optimize processing effi￾ciency, we batch each statement by token length. This approach minimizes the total padding required, ensuring that statements with similar padding are grouped within the same batch. Condition Preprocessing Every condition in our dataset initially starts with the… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: AEN runtime architecture for constant data evaluation 3.2.2 Training Methodology We train on a NVIDIA RTX 4090. Data points, Learning Rate and Batch size, all varied from 50k to 5M points, 1e-5 to 1e-6 and 64 to 256 respectively. We employ the all-MiniLM-L6-v2 and all-…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 6 canonical work pages

  1. [1]

    Comprehensive exploration of synthetic data generation: A survey, 2024

    André Bauer, Simon Trapp, Michael Stenger, Robert Leppich, Samuel Kounev, Mark Leznik, Kyle Chard, and Ian Foster. Comprehensive exploration of synthetic data generation: A survey, 2024. URL https://arxiv.org/abs/2401.02524

    arXiv 2024

  2. [2]

    Llm2vec: Large language models are secretly powerful text encoders, 2024

    Parishad BehnamGhader, Vaibhav Adlakha, Marius Mosbach, Dzmitry Bahdanau, Nicolas Chapados, and Siva Reddy. Llm2vec: Large language models are secretly powerful text encoders, 2024. URL https://arxiv.org/abs/2404.05961

    arXiv 2024

  3. [3]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin...

    arXiv 2020

  4. [4]

    John, Noah Constant, Mario Guajardo-Cespedes, Steve Yuan, Chris Tar, Yun-Hsuan Sung, Brian Strope, and Ray Kurzweil

    Daniel Cer, Yinfei Yang, Sheng yi Kong, Nan Hua, Nicole Limtiaco, Rhomni St. John, Noah Constant, Mario Guajardo-Cespedes, Steve Yuan, Chris Tar, Yun-Hsuan Sung, Brian Strope, and Ray Kurzweil. Universal sentence encoder, 2018. URL https://arxiv.org/abs/1803.11175

    arXiv 2018

  5. [5]

    A tutorial on kernel density estimation and recent advances, 2017

    Yen-Chi Chen. A tutorial on kernel density estimation and recent advances, 2017. URL https://arxiv.org/abs/1704.03924

    arXiv 2017

  6. [6]

    Moo K. Chung. The heat kernel and gaussian kernel. Lecture notes, University of Wisconsin-Madison, 2004. URL https://pages.stat.wisc.edu/ mchung/teaching/MIA/reading/diffusion.gaussian.kernel.pdf

    work page 2004

  7. [7]

    Supervised learning of universal sentence representations from natural language inference data, 2018

    Alexis Conneau, Douwe Kiela, Holger Schwenk, Loic Barrault, and Antoine Bordes. Supervised learning of universal sentence representations from natural language inference data, 2018. URL https://arxiv.org/abs/1705.02364

    arXiv 2018

  8. [8]

    Bert: Pre-training of deep bidirectional transformers for language understanding, 2019

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding, 2019. URL https://arxiv.org/abs/1810.04805

    arXiv 2019

Show all 28 references
  1. [9]

    Learning thematic similarity metric from article sections using triplet networks

    Liat Ein Dor, Yosi Mass, Alon Halfon, Elad Venezian, Ilya Shnayderman, Ranit Aharonov, and Noam Slonim. Learning thematic similarity metric from article sections using triplet networks. In Iryna Gurevych and Yusuke Miyao, editors, Proceedings of the 56th Annual Meeting of the ...

    2018 doi

  2. [10]

    V. A. Epanechnikov. Non-parametric estimation of a multivariate probability density. Theory of Probability & Its Applications, 14 0 (1): 0 153--158, 1969

    1969

  3. [11]

    Deep Learning

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning. MIT Press, 2016. http://www.deeplearningbook.org

    2016

  4. [12]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models, 2021. URL https://arxiv.org/abs/2106.09685

    2021 arXiv

  5. [13]

    Sentimentgpt: Exploiting gpt for advanced sentiment analysis and its departure from current machine learning, 2023

    Kiana Kheiri and Hamid Karimi. Sentimentgpt: Exploiting gpt for advanced sentiment analysis and its departure from current machine learning, 2023. URL https://arxiv.org/abs/2307.10234

    2023 arXiv

  6. [14]

    Small language models are good too: An empirical study of zero-shot classification, 2024

    Pierre Lepagnol, Thomas Gerald, Sahar Ghannay, Christophe Servan, and Sophie Rosset. Small language models are good too: An empirical study of zero-shot classification, 2024. URL https://arxiv.org/abs/2404.11122

    2024 arXiv

  7. [15]

    Riva Shalom, and Michal Chalamish

    Avivit Levy, B. Riva Shalom, and Michal Chalamish. A guide to similarity measures, 2024. URL https://arxiv.org/abs/2408.07706

    2024 arXiv

  8. [16]

    Textbooks are all you need ii: phi-1.5 technical report, 2023

    Yuanzhi Li, Sébastien Bubeck, Ronen Eldan, Allie Del Giorno, Suriya Gunasekar, and Yin Tat Lee. Textbooks are all you need ii: phi-1.5 technical report, 2023. URL https://arxiv.org/abs/2309.05463

    2023 arXiv

  9. [17]

    When ai makes ai: Synthetic data, model distillation, and model collapse, 2024

    Scott Martens. When ai makes ai: Synthetic data, model distillation, and model collapse, 2024. URL https://jina.ai/news/when-ai-makes-ai -synthetic-data-model-distillation -and-model-collapse/

    2024

  10. [18]

    A comprehensive overview of large language models, 2024

    Humza Naveed, Asad Ullah Khan, Shi Qiu, Muhammad Saqib, Saeed Anwar, Muhammad Usman, Naveed Akhtar, Nick Barnes, and Ajmal Mian. A comprehensive overview of large language models, 2024. URL https://arxiv.org/abs/2307.06435

    2024 arXiv

  11. [19]

    Generalized mean shift with triangular kernel profile, 2020

    Sébastien Razakarivony and Axel Barrau. Generalized mean shift with triangular kernel profile, 2020. URL https://arxiv.org/abs/2001.02165

    2020 arXiv

  12. [20]

    Sentence-bert: Sentence embeddings using siamese bert-networks, 2019

    Nils Reimers and Iryna Gurevych. Sentence-bert: Sentence embeddings using siamese bert-networks, 2019. URL https://arxiv.org/abs/1908.10084

    2019 arXiv

  13. [21]

    A sharp fourier inequality and the Epanechnikov kernel

    Sean Richardson. A sharp fourier inequality and the Epanechnikov kernel. arXiv preprint arXiv:2310.09713, 2023

    2023 arXiv

  14. [22]

    Facenet: A unified embedding for face recognition and clustering

    Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A unified embedding for face recognition and clustering. In 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, June 2015. doi:10.1109/cvpr.2015.7298682. URL http://dx.doi.org/10.1109/C...

    2015

  15. [23]

    Augmented sbert: Data augmentation method for improving bi-encoders for pairwise sentence scoring tasks

    Nandan Thakur, Nils Reimers, Johannes Daxenberger, and Iryna Gurevych. Augmented sbert: Data augmentation method for improving bi-encoders for pairwise sentence scoring tasks. arXiv preprint arXiv:2010.08240, 10 2020. URL https://arxiv.org/abs/2010.08240

    2010 arXiv

  16. [24]

    Efficient few-shot learning without prompts, 2022

    Lewis Tunstall, Nils Reimers, Unso Eun Seo Jo, Luke Bates, Daniel Korat, Moshe Wasserblat, and Oren Pereg. Efficient few-shot learning without prompts, 2022. URL https://arxiv.org/abs/2209.11055

    2022 arXiv

  17. [25]

    Generating faithful synthetic data with large language models: A case study in computational social science, 2023

    Veniamin Veselovsky, Manoel Horta Ribeiro, Akhil Arora, Martin Josifoski, Ashton Anderson, and Robert West. Generating faithful synthetic data with large language models: A case study in computational social science, 2023. URL https://arxiv.org/abs/2305.15041

    2023 arXiv

  18. [26]

    Gpt-ner: Named entity recognition via large language models, 2023

    Shuhe Wang, Xiaofei Sun, Xiaoya Li, Rongbin Ouyang, Fei Wu, Tianwei Zhang, Jiwei Li, and Guoyin Wang. Gpt-ner: Named entity recognition via large language models, 2023. URL https://arxiv.org/abs/2304.10428

    2023 arXiv

  19. [27]

    Aggarwal, Jian Pei, and Yuanchun Zhou

    Zaitian Wang, Pengfei Wang, Kunpeng Liu, Pengyang Wang, Yanjie Fu, Chang-Tien Lu, Charu C. Aggarwal, Jian Pei, and Yuanchun Zhou. A comprehensive survey on data augmentation, 2024. URL https://arxiv.org/abs/2405.09591

    2024

  20. [28]

    Wells and Kai Ming Ting

    Jonathan R. Wells and Kai Ming Ting. A simple efficient density estimator that enables fast systematic search, 2017. URL https://arxiv.org/abs/1707.00783

    2017 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.