Pith. sign in

REVIEW 4 major objections 6 minor 39 references

Latte: Transfering LLMs` Latent-level Knowledge for Few-shot Tabular Learning

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Latte transfers an LLM's pooled hidden-state vector into a small tabular model through a KL-aligned attention query and reports consistent few-shot gains.

desk verdict The core training-time LLM-knowledge-transfer idea is sound, but the paper's own Table 2 contradicts its SOTA claim, so the results should be treated as unverified. read the letter →

arxiv 2505.05237 v1 pith:L72PW4ST submitted 2025-05-08 cs.LG

Ruxue Shi , Hengrui Gu , Hangting Ye , Yiwei Dai , Xu Shen , Xin Wang This is my paper · ORCID
classification cs.LG
keywords few-shottabularlearninglatent-levelknowledgetransferLLMdistillationadaptersemanticencoderunsupervisedmeta-learningclassificationregression
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 introduces Latte, a training-time framework for few-shot tabular learning that treats a large language model as a teacher rather than as a test-time predictor. Its claim is that a single average-pooled vector of the LLM's last-layer hidden states, computed from a text prompt describing the task and its features, carries task-relevant prior knowledge that can be distilled into a small downstream model. That distillation works through a knowledge adapter that uses a KL divergence loss to reshape a query vector, which then attends over semantic feature embeddings to produce a task-aware row representation. The paper reports that this consistently beats prior text-rule feature engineering, with an average 4.22% AUC improvement over FeatLLM across shot settings, while calling the LLM only once per dataset. A reader should care because it points to a cheap way to use LLM priors: harvest knowledge once, offline, and run a small model at inference.

What carries the argument

The load-bearing object is the latent knowledge vector $h_{\mathcal{M}}$, obtained by averaging the last-layer hidden states of an LLM over a metadata prompt (Eq. 4). This vector is distilled into a global query $q$ via a KL divergence loss (Eq. 6) inside a knowledge adapter, and the resulting query $q_{\text{LLM}}$ attends over feature embeddings produced by a semantic-aware tabular encoder. The encoder itself encodes each feature value using BERT: categorical values as pooled encodings of feature-name-plus-value text, numerical values as the feature-name embedding multiplied by the scalar value. The mechanism's job is to let the LLM's prior reweight which feature values matter, while a constant $\eta$ blends the LLM-guided representation with a general [CLS] representation. All of this is trained first on pseudo-labeled clusters from unlabeled rows, then on the few labeled examples.

What would settle it

Scramble or replace the metadata prompt with an irrelevant one and rerun Latte on the same datasets; if AUC and MSE do not clearly drop, the claimed transfer from the LLM is not happening. A sharper check is to train a linear probe on the pooled hidden-state vector to predict the target label: chance-level probe accuracy would show the vector carries no task signal.

Watch

Extended reading notes

Core claim

Latte's central discovery is that the hidden states of an LLM, hooked at the last transformer layer and pooled over a metadata-only prompt, can act as a latent prior for a downstream tabular model. The paper claims that, unlike text-level rules generated autoregressively, this latent vector is more informative and less prone to hallucination, and it can be transferred by aligning the LLM's vector with a query vector produced by a GTransformer, then using attention over semantically encoded feature values. Combined with an unsupervised meta-learning stage that clusters unlabeled rows into pseudo-labeled N-way K-shot tasks, this enables both classification and regression from very few labeled samples. The paper reports consistent gains over ten baselines, including a 4.22% average improvement over the strongest text-engineering baseline, and shows the learned representations separate classes even with four labeled samples.

Load-bearing premise

The load-bearing assumption is that one summary vector from the LLM's final layer, computed from a text description of the task and its features, contains useful task knowledge that a learning objective can press into a small model; if that vector is mostly prompt-formatting noise, the whole transfer step adds nothing.

Editorial extensions

If this is right

  • Because the LLM is invoked once at preprocessing time rather than once per test row, deployment latency and inference cost for Latte are independent of the number of test samples.
  • The same pipeline handles regression without modification, whereas several competing few-shot methods are classification-only.
  • Using unlabeled rows through clustering pseudo-labels extends the effective supervision beyond the labeled set, which matters when only a handful of labels exist.
  • The reported 4.22% average AUC gain over the best text-level method implies that latent-vector distillation can substitute for generated textual rules in feature engineering.

Reading between the lines

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

  • Editorial inference: the claim that a single pooled vector suffices could be tested by ablating the pooling step, replacing average-pooling with attention pooling or a [CLS]-style token to see whether the precise aggregation matters or any summarization works.
  • Editorial inference: because the prompt contains only metadata, the same recipe could be applied to wide or high-cardinality tables where serializing entire rows exceeds the LLM context window.
  • Editorial inference: the constant $\eta$ that blends LLM-guided and general representations is fixed; a task-adaptive or learned $\eta$ might be needed for domains where the LLM's priors are weak or outdated, and this is a natural extension the paper does not explore.
  • Editorial inference: probing the pooled hidden-state vector with a linear classifier before training would measure how much task signal it carries independently of downstream accuracy.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. Latte proposes a training-time framework for few-shot tabular learning. For a given tabular task, it feeds the dataset metadata (task and feature descriptions) into an LLM, average-pools the last-layer hidden states to obtain a 'latent knowledge' vector, and uses this vector to guide a downstream transformer-based tabular encoder through a knowledge adapter and KL-style losses. The method also includes an unsupervised meta-learning stage that clusters unlabeled data to generate pseudo-labels. Experiments are reported on nine datasets (six classification, three regression) against ten baselines, with the headline claim that Latte consistently outperforms all competing methods and exceeds FeatLLM at every shot setting.

Significance. If the results were as claimed, the paper would make a useful contribution: it avoids per-sample LLM calls at inference (one LLM call per task), combines LLM prior knowledge with unlabeled-data semi-supervision, and handles regression without architectural modification. The authors also provide code. However, the central empirical claim is contradicted by the paper's own Table 2, so the significance as stated is not established.

major comments (4)
  1. [§4.2, Table 2] The sentence 'our method exceeds the performance of the state-of-the-art method FeatLLM in all shot settings' is contradicted by Table 2. On Diabetes, FeatLLM outperforms Latte at all five shot counts (80.28 vs 72.06 at 4 shots; 79.38 vs 73.70 at 8; 80.15 vs 76.78 at 16; 80.06 vs 77.01 at 32; 80.91 vs 78.32 at 64), and on Blood at 8 shots FeatLLM scores 70.37 vs Latte's 69.97. These are at least six of the thirty Latte-versus-FeatLLM comparisons in the classification table, so 'consistently outperforms all competing methods' is false as stated. The reported average improvement of 4.22% is not reproducible by straightforward aggregation of the classification results shown in Table 2. Because the abstract, Section 1, and Section 5 all rely on this claim, the paper's main conclusion is unsupported.
  2. [§3.4, Eqs. (6) and (11)] Equation (6) defines qLLM = KL(W0 hM/τ, q/τ), and Eq. (11) uses the same expression LKL = KL(W0 hM/τ, q/τ) as a scalar loss. This is not a well-defined use of KL divergence: KL divergence is defined between probability distributions, whereas W0 hM and q are raw real-valued vectors and no softmax or distributional normalization is specified. As written, Eq. (6) cannot produce a vector qLLM under the standard definition of KL, and Eq. (11) cannot be computed. The authors should specify the actual objective (for example, KL after converting both vectors to distributions, or an MSE/cosine alignment loss) and use consistent notation. Since this loss is the mechanism by which LLM knowledge is transferred, the current formulation is not reproducible.
  3. [§4.3, Table 3] The ablation study is conducted on only the Heart dataset, so it does not support the general claim that each component is crucial across datasets and task types. In addition, the table reports no significance tests, and several compared configurations are within one standard deviation of the full model (for example, at 4 shots the full model is 86.10±5.42 vs 85.16±4.44 for the configuration without the LLM-knowledge and meta components). The text's assertion that 'in all cases, modifying any of the ablated components leads to a decline in performance' is therefore stronger than the evidence provided.
  4. [§3.5, Eq. (10)] The pre-training stage depends on clustering unlabeled data with k centroids (Eq. 10), and the text then says the procedure 'randomly select k samples from each cluster to create an N-way K-shot meta-training task.' The relationship between k, N, and K is never specified, and k appears both as the number of clusters and as the number of samples selected per cluster. Without this detail the unsupervised meta-learning procedure is not reproducible, and the dependence of the method on the choice of k is not analyzed.
minor comments (6)
  1. [Title] The title contains a typo: 'Transfering' should be 'Transferring'.
  2. [Table 2] The dataset name is misspelled as 'Boold' in the header; it should be 'Blood'.
  3. [Table 4] The word 'Totle' in the table footer should be 'Total'.
  4. [References] References [Han et al., 2024a] and [Han et al., 2024b] point to the same arXiv paper and should be consolidated.
  5. [§3.3 vs §4.1] Section 3.3 says the knowledge vector is obtained from 'the last transformer layer,' but the implementation details say 'The activation vector in the LLMs is obtained from the 30 layers.' If LLaMA2-7B has 32 layers, these statements conflict and should be reconciled.
  6. [§4.1] The sentence 'we evaluate the proposed model against with baseline' is ungrammatical; it should read 'against baseline methods' or similar.

Circularity Check

1 steps flagged · score 2.0 of 10

No load-bearing circularity in the main derivation; one interpretability claim is by-construction confirmation of the KL loss.

  1. fitted input called prediction [Section 4.5, Eqs. (6), (11), (12)]
    "To distill task-related semantic knowledge from the LLM into the global query vector q and obtain the task-relevant global query vector qLLM, we apply the following knowledge distillation formula: qLLM = KL (W0hM/τ,q/τ ) ... The results of the heatmap demonstrate that our model, after meta-learning on unlabeled data, captures task-relevant semantic information. For example, the representations learned by Latte exhibit high semantic similarity with 'patient' and 'heart disease,' indicating its effective learning of task-relevant semantic knowledge."

    The training losses are L_meta = L_KL + L_pseudo and L_pre = L_KL + L_true, with L_KL = KL(W0hM/τ, q/τ). This objective directly forces the model's query representation q toward the LLM's pooled hidden state hM. Therefore the Section 4.5 heatmap, which reports high semantic similarity between the learned representations and LLM activations and presents it as evidence that Latte captures task-relevant semantic information, is a direct readout of the optimized loss term rather than an independent discovery. The similarity is high by construction whenever the KL term converges. This does not affect the held-out benchmark comparisons, but it makes the interpretability claim a fitted objective reported as empirical evidence.

full rationale

The central derivation is not circular. The LLM target hM = Average([h1,h2,...]) is an external, fixed vector obtained from metadata-only prompts before training; the student query q is trained toward hM via L_KL, while labels or cluster pseudo-labels provide the remaining supervision. Final predictions are functions of tabular features and this fixed target, evaluated on held-out data; no equation fits a parameter to the evaluation set and then renames it a prediction. Pseudo-labels come from clustering unlabeled rows (Eq. 10), not from test labels or the model's own parameters. There are no author self-citations used as load-bearing evidence, and the claim that latent states are more informative than text states rests on external references, not on a uniqueness theorem of the authors. The only by-construction element is the Section 4.5 semantic-similarity heatmap: because L_KL directly minimizes divergence between q and hM, high similarity between learned representations and LLM activations is the optimized training objective, not an independent validation. This is a minor, non-central circularity. Separately, the paper's headline statement that Latte 'exceeds FeatLLM in all shot settings' is contradicted by its own Table 2, for example on Diabetes at every shot, but that is an internal-consistency or correctness issue rather than a derivational circularity, so it does not raise the circularity score.

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

The central claim rests on the assumption that LLM hidden states contain task-relevant knowledge and that the distillation target is informative. Several hyperparameters (eta, tau, k, L) are free constants. No new physical entities are introduced.

free parameters (5)
  • eta
    Mixing weight between LLM-guided representation h_LLM and the CLS token in Eq. (9); a constant chosen without a stated value or validation.
  • tau
    Temperature in the KL divergence in Eq. (6), not specified.
  • k (cluster count)
    Number of centroids in Eq. (10) for pseudo-label generation; not specified.
  • LLM activation layer L = 30
    Layer from which hidden states are hooked; the paper studies L in Figure 2 and uses 30 for the main results, implying per-task tuning.
  • noise m
    Random corruption added to features in Eq. (10); distribution not specified.
assumptions (5)
  • domain assumption The last-layer average-pooled hidden state of an LLM over a metadata prompt encodes task-relevant prior knowledge.
    Central premise of the knowledge adapter (Section 3.3, Eq. 3-4).
  • domain assumption Latent-level knowledge is more informative than text-level knowledge for this task.
    Justifies design; supported by citations [Zhou et al., 2024; Chen et al., 2024] but not independently tested for tabular learning in this paper.
  • domain assumption BERT embeddings of feature names and values provide useful semantic representations.
    Underlies the semantic-aware encoder (Eq. 1).
  • domain assumption Clustering on unlabeled data yields pseudo-labels accurate enough for meta-learning.
    Used in pre-training (Eq. 10-11).
  • ad hoc to paper The KL distillation loss L_KL can align the model's query with the LLM hidden state without degrading task performance.
    A design assumption of the knowledge adapter (Eq. 6).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Latte: Transfering LLMs` Latent-level Knowledge for Few-shot Tabular Learning." pith.science (2026). https://pith.science/paper/L72PW4ST

@misc{pith2026250505237,
  author       = {Pith},
  title        = {Pith review of: Latte: Transfering LLMs` Latent-level Knowledge for Few-shot Tabular Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L72PW4ST}},
  note         = {Machine review of arXiv:2505.05237}
}
read the original abstract

Few-shot tabular learning, in which machine learning models are trained with a limited amount of labeled data, provides a cost-effective approach to addressing real-world challenges. The advent of Large Language Models (LLMs) has sparked interest in leveraging their pre-trained knowledge for few-shot tabular learning. Despite promising results, existing approaches either rely on test-time knowledge extraction, which introduces undesirable latency, or text-level knowledge, which leads to unreliable feature engineering. To overcome these limitations, we propose Latte, a training-time knowledge extraction framework that transfers the latent prior knowledge within LLMs to optimize a more generalized downstream model. Latte enables general knowledge-guided downstream tabular learning, facilitating the weighted fusion of information across different feature values while reducing the risk of overfitting to limited labeled data. Furthermore, Latte is compatible with existing unsupervised pre-training paradigms and effectively utilizes available unlabeled samples to overcome the performance limitations imposed by an extremely small labeled dataset. Extensive experiments on various few-shot tabular learning benchmarks demonstrate the superior performance of Latte, establishing it as a state-of-the-art approach in this domain

Figures

Figures reproduced from arXiv: 2505.05237 by the authors.

Figure 1
Figure 1. Overview of our proposed Latte. This framework begins with the extraction of task-relevant knowledge from LLMs. Knowledge adapters are then employed to guide pre-training on unlabeled data, followed by semantic-aware fine-tuning using few-shot labeled samples. Here we refer to the unsupervised pre-training and the semantic-aware fine-tuning as StageI and StageII, respectively. 3 Method In this section, we innovative… view at source ↗
Figure 2
Figure 2. Latte’s parameter experiments and representation visualizations. (a-b) Impact of the number of labeled samples and LLM activation layers on Latte’s performance in classification and regression tasks. (c-e) Visualization of Latte’s learned representations: (c) original representations, and Latte representations trained on (d) 4 and (e) 64 labeled samples. Answer the question Does the coronary angiography of this pati… view at source ↗
Figure 3
Figure 3. The heatmap visualizes the task-related semantic informa [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 21 canonical work pages

  1. [1]

    Tabnet: Attentive interpretable tabular learning

    [Arik and Pfister, 2021] Sercan ¨O Arik and Tomas Pfister. Tabnet: Attentive interpretable tabular learning. In Pro- ceedings of the AAAI conference on artificial intelligence, volume 35, pages 6679–6687,

    work page 2021

  2. [4]

    Random forests

    [Breiman, 2001] Leo Breiman. Random forests. Machine learning, 45:5–32,

    2001

  3. [11]

    Revisiting deep learning models for tabular data

    [Gorishniy et al., 2021] Yury Gorishniy, Ivan Rubachev, Valentin Khrulkov, and Artem Babenko. Revisiting deep learning models for tabular data. Advances in Neural In- formation Processing Systems, 34:18932–18943,

    work page 2021

  4. [13]

    Large language models can auto- matically engineer features for few-shot tabular learning

    [Han et al., 2024b] Sungwon Han, Jinsung Yoon, Sercan O Arik, and Tomas Pfister. Large language models can auto- matically engineer features for few-shot tabular learning. arXiv preprint arXiv:2404.09491,

    arXiv

  5. [14]

    Tabllm: Few-shot classification of tab- ular data with large language models

    [Hegselmann et al., 2023] Stefan Hegselmann, Alejandro Buendia, Hunter Lang, Monica Agrawal, Xiaoyi Jiang, and David Sontag. Tabllm: Few-shot classification of tab- ular data with large language models. In International Conference on Artificial Intelligence and Statistics , pages 5549–5581. PMLR,

    work page 2023

  6. [15]

    Tabpfn: A transformer that solves small tabular classification prob- lems in a second

    [Hollmann et al., 2023] Noah Hollmann, Samuel M ¨uller, Katharina Eggensperger, and Frank Hutter. Tabpfn: A transformer that solves small tabular classification prob- lems in a second. In The Eleventh International Confer- ence on Learning Representations,

    work page 2023

  7. [16]

    Tabtransformer: Tabu- lar data modeling using contextual embeddings

    [Huang et al., 2020] Xin Huang, Ashish Khetan, Milan Cvitkovic, and Zohar Karnin. Tabtransformer: Tabu- lar data modeling using contextual embeddings. arXiv preprint arXiv:2012.06678,

    arXiv 2020

  8. [17]

    Lightgbm: A highly efficient gradient boost- ing decision tree

    [Ke et al., 2017] Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. Lightgbm: A highly efficient gradient boost- ing decision tree. Advances in neural information process- ing systems, 30,

    work page 2017

Show all 39 references
  1. [20]

    Self- normalizing neural networks

    [Klambauer et al., 2017] G¨unter Klambauer, Thomas Un- terthiner, Andreas Mayr, and Sepp Hochreiter. Self- normalizing neural networks. Advances in neural infor- mation processing systems, 30,

    2017

  2. [21]

    Logistic regression

    [LaValley, 2008] Michael P LaValley. Logistic regression. Circulation, 117(18):2395–2399,

    2008

  3. [23]

    Classification and regression trees

    [Loh, 2011] Wei-Yin Loh. Classification and regression trees. Wiley interdisciplinary reviews: data mining and knowledge discovery, 1(1):14–23,

    2011

  4. [25]

    Tadam: Task dependent adaptive metric for improved few-shot learning

    [Oreshkin et al., 2018] Boris Oreshkin, Pau Rodr´ıguez L ´opez, and Alexandre Lacoste. Tadam: Task dependent adaptive metric for improved few-shot learning. Advances in neural information processing systems, 31,

    2018

  5. [26]

    Hallucinations in llms: Understand- ing and addressing challenges

    [Perkovi´c et al., 2024] Gabrijela Perkovi´c, Antun Drobnjak, and Ivica Boti ˇcki. Hallucinations in llms: Understand- ing and addressing challenges. In 2024 47th MIPRO ICT and Electronics Convention (MIPRO), pages 2084–2088. IEEE,

    2024

  6. [27]

    Neural oblivious decision ensem- bles for deep learning on tabular data

    [Popov et al., 2019] Sergei Popov, Stanislav Morozov, and Artem Babenko. Neural oblivious decision ensem- bles for deep learning on tabular data. arXiv preprint arXiv:1909.06312,

    2019 arXiv

  7. [28]

    Catboost: unbiased boosting with categor- ical features

    [Prokhorenkova et al., 2018] Liudmila Prokhorenkova, Gleb Gusev, Aleksandr V orobev, Anna Veronika Dorogush, and Andrey Gulin. Catboost: unbiased boosting with categor- ical features. Advances in neural information processing systems, 31,

    2018

  8. [29]

    Machine learning in healthcare: A review

    [Shailaja et al., 2018] K Shailaja, Banoth Seetharamulu, and MA Jabbar. Machine learning in healthcare: A review. In 2018 Second international conference on electronics, communication and aerospace technology (ICECA), pages 910–914. IEEE,

    2018

  9. [30]

    Tablet: Learning from instructions for tabular data

    [Slack and Singh, 2023] Dylan Slack and Sameer Singh. Tablet: Learning from instructions for tabular data. arXiv preprint arXiv:2304.13188,

    2023 arXiv

  10. [31]

    Prototypical networks for few-shot learning

    [Snell et al., 2017] Jake Snell, Kevin Swersky, and Richard Zemel. Prototypical networks for few-shot learning. Advances in neural information processing systems , 30,

    2017

  11. [32]

    Saint: Improved neural networks for tabular data via row attention and contrastive pre-training

    [Somepalli et al., 2021] Gowthami Somepalli, Micah Gold- blum, Avi Schwarzschild, C Bayan Bruss, and Tom Gold- stein. Saint: Improved neural networks for tabular data via row attention and contrastive pre-training. arXiv preprint arXiv:2106.01342,

    2021 arXiv

  12. [33]

    Autoint: Automatic feature interaction learning via self-attentive neural networks

    [Song et al., 2019] Weiping Song, Chence Shi, Zhiping Xiao, Zhijian Duan, Yewen Xu, Ming Zhang, and Jian Tang. Autoint: Automatic feature interaction learning via self-attentive neural networks. In Proceedings of the 28th ACM international conference on information and knowl- ...

    2019

  13. [34]

    Llama: Open and efficient founda- tion language models

    [Touvron et al., 2023] Hugo Touvron, Thibaut Lavril, Gau- tier Izacard, Xavier Martinet, Marie-Anne Lachaux, Tim- oth´ee Lacroix, Baptiste Rozi`ere, Naman Goyal, Eric Ham- bro, Faisal Azhar, et al. Llama: Open and efficient founda- tion language models. arXiv preprint arXiv:23...

    2023 arXiv

  14. [35]

    Generalizing from a few ex- amples: A survey on few-shot learning

    [Wang et al., 2020] Yaqing Wang, Quanming Yao, James T Kwok, and Lionel M Ni. Generalizing from a few ex- amples: A survey on few-shot learning. ACM computing surveys (csur), 53(3):1–34,

    2020

  15. [36]

    Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems

    [Wang et al., 2021] Ruoxi Wang, Rakesh Shivanna, Derek Cheng, Sagar Jain, Dong Lin, Lichan Hong, and Ed Chi. Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems. In Pro- ceedings of the web conference 2021 , pages 1785–1797,

    2021

  16. [37]

    Emergent abilities of large language models

    [Wei et al., 2022] Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yo- gatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. Emergent abilities of large language models. arXiv preprint arXiv:2206.07682,

    2022 arXiv

  17. [38]

    From supervised to generative: A novel paradigm for tabular deep learning with large lan- guage models

    [Wen et al., 2024] Xumeng Wen, Han Zhang, Shun Zheng, Wei Xu, and Jiang Bian. From supervised to generative: A novel paradigm for tabular deep learning with large lan- guage models. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages...

    2024

  18. [39]

    Vime: Extending the suc- cess of self-and semi-supervised learning to tabular do- main

    [Yoon et al., 2020] Jinsung Yoon, Yao Zhang, James Jordon, and Mihaela van der Schaar. Vime: Extending the suc- cess of self-and semi-supervised learning to tabular do- main. Advances in Neural Information Processing Sys- tems, 33:11033–11043,

    2020

  19. [40]

    How alignment and jailbreak work: Explain llm safety through intermediate hidden states

    [Zhou et al., 2024] Zhenhong Zhou, Haiyang Yu, Xinghua Zhang, Rongwu Xu, Fei Huang, and Yongbin Li. How alignment and jailbreak work: Explain llm safety through intermediate hidden states. arXiv preprint arXiv:2406.05644, 2024

    2024 arXiv

  20. [2001]

    Ai in finance: challenges, techniques, and opportunities

    [Cao, 2022] Longbing Cao. Ai in finance: challenges, techniques, and opportunities. ACM Computing Surveys (CSUR), 55(3):1–38,

    2022

  21. [2008]

    D2r2: Diffusion-based representation with random distance matching for tabular few-shot learn- ing

    [Liu et al., 2024] Ruoxue Liu, Linjiajie Fang, Wenjia Wang, and Bingyi Jing. D2r2: Diffusion-based representation with random distance matching for tabular few-shot learn- ing. In The Thirty-eighth Annual Conference on Neural Information Processing Systems,

    2024

  22. [2011]

    Stunt: Few-shot tab- ular learning with self-generated tasks from unlabeled ta- bles

    [Nam et al., 2023] Jaehyun Nam, Jihoon Tack, Kyungmin Lee, Hankook Lee, and Jinwoo Shin. Stunt: Few-shot tab- ular learning with self-generated tasks from unlabeled ta- bles. In The Eleventh International Conference on Learn- ing Representations,

    2023

  23. [2016]

    A closer look at few-shot classification

    [Chen et al., 2019] Wei-Yu Chen, Yen-Cheng Liu, Zsolt Kira, Yu-Chiang Frank Wang, and Jia-Bin Huang. A closer look at few-shot classification. arXiv preprint arXiv:1904.04232,

    2019 arXiv

  24. [2017]

    Tabnn: A universal neural network solution for tabular data

    [Ke et al., 2018] Guolin Ke, Jia Zhang, Zhenhui Xu, Jiang Bian, and Tie-Yan Liu. Tabnn: A universal neural network solution for tabular data

    2018

  25. [2018]

    Deepgbm: A deep learning frame- work distilled by gbdt for online prediction tasks

    [Ke et al., 2019] Guolin Ke, Zhenhui Xu, Jia Zhang, Jiang Bian, and Tie-Yan Liu. Deepgbm: A deep learning frame- work distilled by gbdt for online prediction tasks. In Pro- ceedings of the 25th ACM SIGKDD International Confer- ence on Knowledge Discovery & Data Mining, pages 384– 394,

    2019

  26. [2019]

    Recontab: Regularized con- trastive representation learning for tabular data

    [Chen et al., 2023] Suiyao Chen, Jing Wu, Naira Hov- akimyan, and Handong Yao. Recontab: Regularized con- trastive representation learning for tabular data. arXiv preprint arXiv:2310.18541,

    2023 arXiv

  27. [2020]

    Scarf: Self-supervised contrastive learn- ing using random feature corruption

    [Bahri et al., 2021] Dara Bahri, Heinrich Jiang, Yi Tay, and Donald Metzler. Scarf: Self-supervised contrastive learn- ing using random feature corruption. arXiv preprint arXiv:2106.15147,

    2021 arXiv

  28. [2021]

    Gradient boosting neural networks: Grownet

    [Badirli et al., 2020] Sarkhan Badirli, Xuanqing Liu, Zheng- ming Xing, Avradeep Bhowmik, Khoa Doan, and Sathiya S Keerthi. Gradient boosting neural networks: Grownet. arXiv preprint arXiv:2002.07971,

    2020 arXiv

  29. [2022]

    Xgboost: A scalable tree boosting system

    [Chen and Guestrin, 2016] Tianqi Chen and Carlos Guestrin. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowl- edge discovery and data mining, pages 785–794,

    2016

  30. [2023]

    In- side: Llms’ internal states retain the power of hallucination detection

    [Chen et al., 2024] Chao Chen, Kai Liu, Ze Chen, Yi Gu, Yue Wu, Mingyuan Tao, Zhihang Fu, and Jieping Ye. In- side: Llms’ internal states retain the power of hallucination detection. arXiv preprint arXiv:2402.03744,

    2024 arXiv

  31. [2024]

    Do llms know about hallucination? an empiri- cal investigation of llm’s hidden states

    [Duan et al., 2024] Hanyu Duan, Yi Yang, and Kar Yan Tam. Do llms know about hallucination? an empiri- cal investigation of llm’s hidden states. arXiv preprint arXiv:2402.09733,

    2024 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.