RegimeNAS: Regime-Aware Differentiable Architecture Search With Theoretical Guarantees for Financial Trading

Prathamesh Devadiga; Yashmitha Shailesh

REVIEW 4 major objections 5 minor 26 references

RegimeNAS: Regime-Aware Differentiable Architecture Search With Theoretical Guarantees for Financial Trading

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read RegimeNAS claims that making architecture search regime-aware—gating volatility, trend, and range blocks by market state—cuts cryptocurrency price prediction error by 80.3% versus the best recurrent baseline.

desk verdict RegimeNAS has a plausible core idea, but its headline test metrics are identical to the search-validation results and Algorithm 1 never evaluates a separate test set, so the 80.3% MAE reduction is likely a selection artifact. read the letter →

arxiv 2508.11338 v1 pith:FII4QUV4 submitted 2025-08-15 cs.LG cs.AI

Prathamesh Devadiga , Yashmitha Shailesh This is my paper · ORCID

classification cs.LGcs.AI

keywords neuralarchitecturesearchmarketregimedetectioncryptocurrencyforecastingBayesianoptimizationmulti-headattentiondynamicgatingLipschitzstabilitymulti-objectiveloss

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

RegimeNAS is a neural architecture search method for cryptocurrency price prediction. Its central claim is that the search should not return one fixed network; instead, the architecture itself should change with the market regime. A multi-head attention module reads multi-timeframe features and outputs regime probabilities; a gating network uses those probabilities to weight three specialized blocks—Volatility, Trend, and Range. Bayesian optimization searches the internals of the blocks, and a multi-objective loss adds volatility matching, output smoothness, and Lipschitz stability. On held-out crypto data the best found architecture reports an 80.3% lower MAE than the best GRU baseline (0.7570 vs 3.8126) with R²=0.9945, and trains in 9 epochs.

What carries the argument

Regime-conditioned gating over specialized blocks. Multi-head self-attention over multi-timeframe features produces regime probabilities $p(r_t)$; an MLP maps these to softmax weights $g_t = [g_V, g_T, g_R]$, and the layer output is $g_V\cdot$V-Block $+ g_T\cdot$T-Block $+ g_R\cdot$R-Block. This makes the effective architecture a function of the detected market state. A Gaussian-process Bayesian optimizer searches the block internals and gating structure, while spectral normalization and Lipschitz regularization—bounds on how much the output can change per unit input change—keep the mapping stable during regime transitions.

What would settle it

Take the final RegimeNAS architecture and replace the regime-conditioned gating weights with random-but-fixed weights drawn from the same distribution; if test MAE stays near 0.757, regime awareness is not carrying the result. Alternatively, compare the attention detector's regime labels to ADX/ATR-based labels on the test set; large disagreement would show the post-hoc regime analysis is not measuring the mechanism the paper claims.

Watch

Extended reading notes

Core claim

The paper's central discovery is that regime awareness can be built directly into architecture search rather than bolted onto a fixed model. RegimeNAS couples a multi-head attention regime detector to a gating MLP that computes soft weights for three specialized blocks; the output is a regime-weighted combination of the blocks. The authors report that this adaptive gating is the largest performance contributor: removing regime detection raises MAE by 63.4% in their ablation. They also provide formal Lipschitz-based bounds guaranteeing that small changes in regime probabilities produce only bounded output changes, and report an 80.3% MAE reduction over the best recurrent baseline on a held-ou

Load-bearing premise

The reported gains are attributed to regime awareness, which assumes the learned regime probabilities actually capture market states, that the gating network uses them meaningfully, and that the baseline comparisons are fair.

Editorial extensions

If this is right

If correct, architecture search for finance should condition on market state rather than return one fixed network; static models leave significant error on the table.
Dynamic gating lets a deployed model switch its computation as regimes shift, and the Lipschitz constraints bound output jumps, which matters for live trading.
The best architecture trains in 9 epochs versus 50+ for recurrent baselines, suggesting adaptive search can be computationally cheaper than standard fixed-model training loops.
The ablation shows removing regime detection increases MAE by 63.4%, making regime awareness the largest single contributor to the reported result.
The block decomposition gives a degree of interpretability: which block is active indicates what kind of market the model believes it is in.

Reading between the lines

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

A testable extension is to compare the learned attention-based regime probabilities against the ADX/ATR thresholds used for the paper's post-hoc regime analysis; agreement would strengthen the causal story, disagreement would suggest the performance gain arises elsewhere.
The same regime-gating design could transfer to other non-stationary series, such as equity indices, volatility indices, or sensor streams, by swapping the financial feature set; the paper only evaluates crypto.
The Lipschitz bound of Eq. 15 can be checked empirically by measuring actual output jumps during detected regime transitions and comparing them to the bound.
Re-running the Bayesian search with different random seeds and data splits would test whether the exact 80.3% MAE reduction is stable or specific to one optimization run.

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.

Desk Editor's Note

read the letter

First, what's actually worth your time: the composition is new. I haven't seen multi-head attention regime detection gating specialized Volatility/Trend/Range blocks inside a Bayesian NAS loop with a multi-objective financial loss. The paper is clearly written and the ablation plan is sensible. That is real, if modest, novelty.

The problems are serious, though. The biggest one: Table I labels RegimeNAS's metrics as held-out test results, but the numbers (Loss=0.5258, MAE=0.7570, RMSE=2.2237, R²=0.9945) are exactly the Generation 4 validation metrics from Table II. Algorithm 1 retrains the best architecture on Dtrain∪Dval and stops; there is no test evaluation step. That is not a minor slip. It means the headline 80.3% reduction over GRU is likely the best-of-search validation score, not a generalization result. The stress-test note on this point holds up on reading the paper, and I'd treat the central empirical claim as unsupported.

Other soft spots, in decreasing order: - The Loss column is internally inconsistent. For RegimeNAS, Loss=0.5258 but RMSE=2.2237, so Loss is not MSE. For XGBoost, Loss=RMSE=153.34. The N-BEATS row is even worse: Loss=0.2750, MAE=452.94, RMSE=533.05, R²=0.9819. The authors flag some of this in a note but don't fix it or explain it. - The theoretical guarantees are not really guarantees. Theorem 1 is an informal restatement of standard BO convergence; Theorem 2 is a trivial Lipschitz composition bound. No proofs, and the paper itself labels theorem 1 as informal. - The method is called differentiable architecture search, but Section III-D uses Bayesian optimization. That's a mislabel. - No code, no data, no error bars, no repeated seeds. The regime analysis uses post-hoc ADX/ATR thresholds, not the learned detector, so the "regime-aware benefit" claim isn't validated against any ground truth. - The authors candidly admit real trading backtests are future work, so the title's "for Financial Trading" is overreach.

Who should read this? Someone studying evaluation failures in NAS or financial ML might use it as a cautionary example. It is not a reliable source for the claimed performance. I would not send it to peer review in its current form; the evaluation issues are load-bearing, not cosmetic. If the authors redo the experiments with a genuinely held-out test set, release code and data, fix the metrics table, and drop the theatrical guarantees, it could become a decent workshop paper. As it stands, desk reject is the right call.

Referee Report

4 major / 5 minor

Summary. RegimeNAS proposes a regime-aware neural architecture search framework for cryptocurrency return prediction. The pipeline combines multi-timeframe multi-head attention for regime detection, a gating MLP that activates specialized volatility/trend/range blocks, Bayesian optimization over an architecture space, and a multi-objective loss with Lipschitz-related stability penalties. The paper claims strong empirical results, including an 80.3% MAE reduction over GRU on a held-out test set and R²=0.9945, plus theoretical convergence and transition-stability guarantees. The manuscript also reports ablations and regime-specific analyses. However, a close reading of the algorithm and tables reveals that the test set evaluation is not actually specified, the reported Loss column is internally inconsistent with the RMSE values, and the theoretical 'guarantees' are informal rather than proven.

Significance. The problem addressed, adaptive architecture search for non-stationary financial markets, is of genuine interest, and the proposed combination of regime detection, gating, specialized blocks, and stability penalties is a plausible design direction. If the empirical claims were properly validated, the paper could be useful to the financial ML community. Unfortunately, the manuscript does not provide machine-checked proofs, code, or data; the theoretical statements are informal; and the central empirical evidence appears to contain a test/validation conflation and metric inconsistencies. As it stands, the results do not support the stated contributions, so the significance is currently limited.

major comments (4)

[§V-A/V-B, Algorithm 1, Tables I-II] The headline 'held-out test set' result is unsupported. Table I's RegimeNAS row (Loss 0.5258, MAE 0.7570, RMSE 2.2237, R² 0.9945) is numerically identical to the 'Generation 4 (Overall Best)' validation row in Table II. Algorithm 1 takes only Dtrain and Dval, retrains the best architecture on Dtrain∪Dval (line 29), and returns; nowhere is a separate Dtest used to compute final metrics. The note under Table I is also self-contradictory: it says metrics 'are reported on the validation set' but 'final comparison uses the held-out test set.' Without a verification that the test set is never used for selection, the claimed 80.3% MAE reduction may be a selection artifact rather than generalization.
[§V-B, Table I] The 'Loss' column is not a well-defined metric. In a squared-error regression framework, Loss should be comparable to RMSE². For RegimeNAS, Loss=0.5258 but RMSE²=4.9446; for GRU, Loss=3.3961 but RMSE²=118.7; for XGBoost, Loss=RMSE=153.34 exactly. These inconsistencies make the Loss column, and the use of 'best Val Loss' in Table II for architecture selection, uninterpretable. The N-BEATS row (MAE=452.94, RMSE=533.05, R²=0.9819) is implausible on a normalized target; the paper's caveat ('requires careful interpretation') does not resolve the contradiction, and D-PAD's low R² (0.6837) likewise suggests an evaluation artifact. The quantitative comparison is therefore not reliable.
[§V-C/V-D, Tables III-IV] There are no repeated runs, error bars, or statistical significance tests; 'fixed random seeds' in §IV-D does not establish variance. More importantly, the attribution of gains to regime awareness is not validated. The post-hoc regime classification in §V-D uses ADX/ATR thresholds, not the learned multi-head attention detector, so the paper never checks whether the learned regime probabilities correspond to the regimes that supposedly drive performance. The ablation claim that disabling regime detection 'increased MAE by 63.4%' (Table III) is therefore confounded with model capacity and evaluation artifacts, and cannot be regarded as evidence for the central mechanism.
[§IV-B, Eq. (15), §IV-C] The claimed 'regime transition stability guarantee' in Theorem 2 is not a theorem. Eq. (15) bounds the output change only in terms of an unspecified L_eff. For the gated output Output_t = Σ g_i f_i(x_t), a change in p while x is fixed gives Σ Δg_i f_i(x_t), whose norm is bounded by ||Δg|| · max_i ||f_i(x_t)||, not by a Lipschitz constant of the gating network alone; no bound on the block outputs is given. The practical mechanisms in §IV-C (spectral normalization, gradient clipping, Lipschitz regularization) are heuristics, and no proof shows that they enforce the stated bound. Theorem 1 is likewise a generic Bayesian-regret statement not tailored to the regime-conditioned search space in Algorithm 1. Thus the 'theoretical guarantees' advertised in the abstract and introduction are not established.

minor comments (5)

[Title/Abstract vs §III-D] The paper is titled 'Differentiable Architecture Search' and the abstract repeats this, but §III-D describes Bayesian optimization over discrete architecture choices with a Gaussian-process surrogate, not differentiable architecture search in the DARTS sense. Please clarify what 'differentiable' refers to (e.g., the gating mechanism only) to avoid mischaracterizing the method.
[§III-A] The dataset description says 'over 20 major cryptocurrencies' from CoinMarketCap (2013-2021), but the exact list of coins, handling of missing data, and the target variable (e.g., raw price vs. log return) are not specified. This makes the N-BEATS/D-PAD anomalies impossible to diagnose.
[§III-F/§IV-D] The loss weights (w_p=1.0, w_v=0.1, w_r=0.05, w_s=0.01) are fixed without any sensitivity analysis or description of how they were chosen; no justification is given. A small study varying these weights would help assess robustness.
[§V-C/Table III] The prose in §V-C is truncated ('22.0', '12.4' without units or percent signs). Also the ablation table reports only MAE/RMSE/R²; adding standard deviations or per-run ranges would be necessary for meaningful comparison.
[§V-D/Table IV] The regime-specific results in Table IV lack sample sizes, time coverage, and any indication of how many periods fall into each regime. The table formatting also appears broken. Please report the number of observations per regime.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported 'held-out test' metrics for RegimeNAS are identical to the validation metrics used for architecture selection, so the headline 80.3% MAE reduction is the selection objective, not an independent test result.

fitted input called prediction [Table I vs Table II; Section V-B; Algorithm 1 (lines 28-30)]
"Table I: 'RegimeNAS (Best Found Arch. - Gen 4) 0.5258 0.7570 2.2237 0.9945'; Table II: '4 (Overall Best) 0.5258 0.7570 2.2237 0.9945 9'; Algorithm 1: '28: Identify best architecture α∗ = arg maxα evaluated f (α) 29: Retrain α∗ on Dtrain ∪ Dval until convergence to get final weights W∗ 30: return α∗, W∗'"

The paper claims Table I reports 'held-out test set' results, but the RegimeNAS row is numerically identical to the Generation-4 validation row in Table II. Algorithm 1's final step retrains on Dtrain∪Dval and returns weights without any evaluation on a separate test split. The reported 'test' MAE/RMSE/R² are therefore the very validation metrics that the Bayesian optimizer used to select α*. The headline 80.3% MAE reduction versus GRU is the selection objective, not an independently measured held-out prediction; it is forced by construction from the validation scores.

full rationale

The paper's theoretical claims (BO convergence, Lipschitz stability) are standard mathematical statements, not circular: Theorem 1 invokes known BO convergence results, and Theorem 2 is a direct Lipschitz composition bound. There is no self-citation chain or imported uniqueness theorem. The NAS architecture itself is a normal optimization loop. However, the central empirical contribution—the claimed 80.3% test MAE reduction and R²=0.9945—is not self-contained: the exact RegimeNAS numbers in Table I ('test') are identical to the Generation-4 'validation' numbers in Table II, and Algorithm 1 contains no held-out test evaluation step. The final model is retrained on Dtrain∪Dval and returned, so the reported test metrics reduce to the validation scores used for architecture selection. This is a fitted-input-called-prediction pattern: the best architecture was chosen by these very scores, and the same scores are then presented as independent held-out performance. This does not necessarily invalidate the method's potential, but it makes the headline empirical claim statistically forced rather than independently verified. The regime-specific analysis in Section V-D further uses ADX/ATR thresholds rather than the learned attention detector, but that is a validation gap rather than a circular reduction. Overall, the derivation chain is substantially compromised at the empirical level, meriting a score of 6.

Assumptions & free parameters 3 free parameters · 5 assumptions · 4 invented entities

The central empirical claim rests on several fitted hyperparameters and unverified assumptions: the multi-objective loss weights are set by hand, the regime thresholds for analysis are chosen post-hoc, the regime detector is not validated against ground truth, and the Lipschitz stability relies on spectral normalization without computing actual constants. The claimed theoretical guarantees are standard BO and Lipschitz composition results, not new derivations.

free parameters (3)

Multi-objective loss weights = w_v=0.1, w_r=0.05, w_s=0.01
Set via preliminary experiments (Sec IV-D); no sensitivity analysis, and they directly shape the searched architecture and final performance.
Regime classification thresholds for evaluation = ADX > 25, ATR > 75th percentile
Used post-hoc in Sec V-D to partition the test set; these are ad hoc and not the same as the learned regime detector, so the regime-specific gains may not reflect the model's own regimes.
Number of regimes and attention dimensions = Nr=3, H=4, dk=dv=64
Fixed choices in Sec IV-D without systematic tuning; the number of regimes affects the gating and the search space.

assumptions (5)

standard math Standard BO convergence (Theorem 1): under regularity assumptions on f(alpha) and kernel, BO asymptotically finds the global optimum.
Invoked in Sec IV-B without proof, relying on [8], [25]. The assumptions are not verified for the actual architecture performance function.
domain assumption Lipschitz continuity of blocks and gating network is ensured by spectral normalization and activation choices, so the bound in Eq. (15) holds.
Sec IV-C describes practical enforcement but does not compute Lipschitz constants; the bound is not empirically verified.
domain assumption The attention-based regime detector outputs probabilities that correspond to meaningful market regimes.
Sec III-B defines p(r_t|X_t) but there is no ground truth validation; the post-hoc analysis uses ADX/ATR instead, suggesting the learned regimes are not trusted.
domain assumption The CoinMarketCap OHLCV dataset is complete and the features are predictive of future returns.
Sec III-A: data not released; no leak checks or feature validity analysis are reported.
ad hoc to paper The 'Loss' column in Table I is a meaningful aggregate loss, despite not matching RMSE^2.
Table I reports Test Loss 0.5258 with RMSE 2.2237; MSE=4.945, so the loss is undefined or miscalculated. This is an auxiliary assumption needed to interpret the reported performance.

invented entities (4)

Volatility Blocks (V-Blocks)
purpose: Handle high-volatility regimes via volatility-gated recurrence and adaptive PLU/Swish activations modulated by sigma_t.
Described qualitatively in Sec III-E1; no standalone evaluation. Ablation shows +22% MAE when removed, but this is within the combined model.
Trend Blocks (T-Blocks)
purpose: Capture trending dynamics via multi-scale temporal convolutions with varying kernel sizes/dilations.
Sec III-E2; no standalone test; ablation +16.8%.
Range Blocks (R-Blocks)
purpose: Model mean-reversion in range-bound markets via attention to a dynamic mean estimate.
Sec III-E3; no standalone test; ablation +7.7%.
Learnable mask M in attention
purpose: Incorporate market-specific biases or relative positional encodings in Eq. (4).
Mentioned in Sec III-B without specification; no analysis of its effect.

how reviews work

0 comments

Cite this review

Pith. "Pith review of RegimeNAS: Regime-Aware Differentiable Architecture Search With Theoretical Guarantees for Financial Trading." pith.science (2026). https://pith.science/paper/FII4QUV4

@misc{pith2026250811338,
  author       = {Pith},
  title        = {Pith review of: RegimeNAS: Regime-Aware Differentiable Architecture Search With Theoretical Guarantees for Financial Trading},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FII4QUV4}},
  note         = {Machine review of arXiv:2508.11338}
}

read the original abstract

We introduce RegimeNAS, a novel differentiable architecture search framework specifically designed to enhance cryptocurrency trading performance by explicitly integrating market regime awareness. Addressing the limitations of static deep learning models in highly dynamic financial environments, RegimeNAS features three core innovations: (1) a theoretically grounded Bayesian search space optimizing architectures with provable convergence properties; (2) specialized, dynamically activated neural modules (Volatility, Trend, and Range blocks) tailored for distinct market conditions; and (3) a multi-objective loss function incorporating market-specific penalties (e.g., volatility matching, transition smoothness) alongside mathematically enforced Lipschitz stability constraints. Regime identification leverages multi-head attention across multiple timeframes for improved accuracy and uncertainty estimation. Rigorous empirical evaluation on extensive real-world cryptocurrency data demonstrates that RegimeNAS significantly outperforms state-of-the-art benchmarks, achieving an 80.3% Mean Absolute Error reduction compared to the best traditional recurrent baseline and converging substantially faster (9 vs. 50+ epochs). Ablation studies and regime-specific analysis confirm the critical contribution of each component, particularly the regime-aware adaptation mechanism. This work underscores the imperative of embedding domain-specific knowledge, such as market regimes, directly within the NAS process to develop robust and adaptive models for challenging financial applications.

Figures

Figures reproduced from arXiv: 2508.11338 by the authors.

**Figure 2.** Distributions of key performance metrics (Top Row: MAE, RMSE, R²) and explored architectural hyperparameters [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 25 canonical work pages

[1]

Deep learning for finance: evidence from algorithmic trading in the cryptocurrency market,

J. B. Heaton, N. G. Polson, and J. H. Witte, “Deep learning for finance: evidence from algorithmic trading in the cryptocurrency market,” Ap- plied Stochastic Models in Business and Industry , vol. 33, no. 1, pp. 3–12, 2017

work page 2017
[2]

Deep learning with long short-term memory networks for financial market predictions,

T. Fischer and C. Krauss, “Deep learning with long short-term memory networks for financial market predictions,” European Journal of Oper- ational Research, vol. 270, no. 2, pp. 654–669, 2018

work page 2018
[3]

Dynamic neural architecture search,

X. Zhang, X. Dai, H. Wang, Y . Chen, D. Lin, Y . Qiao, J. Gao, and Y . He, “Dynamic neural architecture search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 23 599–23 609

work page 2023
[4]

A regime- switching model of long-term stock market returns,

P. Nystrup, B. V . Hansen, H. Madsen, and E. Lindstr ¨om, “A regime- switching model of long-term stock market returns,” The European Journal of Finance , vol. 21, no. 13-14, pp. 1155–1175, 2015

work page 2015
[5]

Attention is all you need,

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in neural information processing systems 30 (NIPS 2017) , 2017, pp. 5998–6008

work page 2017
[6]

KAN: Kolmogorov-Arnold Networks,

Z. Liu, Y . Wang, S. Vaidya, F. Ruehle, J. Halatek, Y . Du, J. Schnei- der, T. Yamamoto, S. Nakatsuji, C. Ho, and M. Tegmark, “KAN: Kolmogorov-Arnold Networks,” arXiv preprint arXiv:2404.19756, 2024

arXiv 2024
[7]

N-BEATS: Neural basis expansion analysis for interpretable time series forecasting,

B. N. Oreshkin, D. Carpov, N. Chapados, and Y . Bengio, “N-BEATS: Neural basis expansion analysis for interpretable time series forecasting,” in International Conference on Learning Representations (ICLR) , 2020

work page 2020
[8]

Neural architecture search with Bayesian optimisation and optimal transport,

K. Kandasamy, W. Neiswanger, J. Schneider, B. P ´oczos, and E. Xing, “Neural architecture search with Bayesian optimisation and optimal transport,” in Advances in Neural Information Processing Systems 31 (NeurIPS 2018), 2018, pp. 2016–2025

work page 2018

Show all 26 references

[9]

BayesNAS: A Bayesian approach for neural architecture search,

H. Zhou, M. Yang, J. Wang, and W. Chen, “BayesNAS: A Bayesian approach for neural architecture search,” in International Conference on Machine Learning (ICML) . PMLR, 2019, pp. 7600–7610

2019
[10]

Automated machine learning: State-of-the-art and open challenges,

B. Chen, Z. Sun, H. Wu, and W. Zheng, “Automated machine learning: State-of-the-art and open challenges,” Neurocomputing, vol. 562, p. 126835, 2023

2023
[11]

Neural architecture search: A survey,

T. Elsken, J. H. Metzen, and F. Hutter, “Neural architecture search: A survey,” Journal of Machine Learning Research , vol. 20, no. 55, pp. 1–21, 2019. [Online]. Available: http://jmlr.org/papers/v20/18-598.html

2019
[12]

DARTS: Differentiable architec- ture search,

H. Liu, K. Simonyan, and Y . Yang, “DARTS: Differentiable architec- ture search,” in International Conference on Learning Representations (ICLR), 2019

2019
[13]

Efficient neural architecture search methods: A survey,

K. Li, C. Li, and Z. Ding, “Efficient neural architecture search methods: A survey,” ACM Computing Surveys , vol. 55, no. 8, pp. 1–35, 2023

2023
[14]

Memory-Efficient Dynamic Neural Architecture Search,

L. Wang, C. Zhang, and J. Yan, “Memory-Efficient Dynamic Neural Architecture Search,” arXiv preprint arXiv:2401.10099 , 2024

2024 arXiv
[15]

Convolutional LSTM network: A machine learning approach for precipitation nowcasting,

S. Xingjian, Z. Chen, H. Wang, D.-Y . Yeung, W.-K. Wong, and W.-c. Woo, “Convolutional LSTM network: A machine learning approach for precipitation nowcasting,” in Advances in neural information processing systems 28 (NIPS 2015) , 2015, pp. 802–810

2015
[16]

D-PAD: A Detail-Preserving Attention-based Decoder for High-Fidelity Time Series Forecasting,

Y . Yuan, L. Shen, X. Ding, Y . Zheng, B. Tang, and C. Xiao, “D-PAD: A Detail-Preserving Attention-based Decoder for High-Fidelity Time Series Forecasting,” arXiv preprint arXiv:2405.13303 , 2024

2024 arXiv
[17]

XGBoost: A scalable tree boosting system,

T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , 2016, pp. 785–794

2016
[18]

Regime switching in international asset allocation,

A. Ang and G. Bekaert, “Regime switching in international asset allocation,” The Journal of Finance, vol. 57, no. 3, pp. 1137–1187, 2002

2002
[19]

Regime-switching models,

J. D. Hamilton, “Regime-switching models,” in The new Palgrave dictionary of economics , vol. 2. Palgrave Macmillan London, 2008, pp. 1–6

2008
[20]

Deep learning with event embedding for stock market prediction,

X. Ding, Y . Zhang, T. Liu, and J. Duan, “Deep learning with event embedding for stock market prediction,” in Proceedings of the 24th ACM international on conference on information and knowledge management, 2015, pp. 1491–1494

2015
[21]

A hidden Markov model for detecting stock market regimes,

M.-J. Kim, M.-K. Lee, and J.-H. Ahn, “A hidden Markov model for detecting stock market regimes,” Applied Intelligence , vol. 49, no. 3, pp. 863–875, 2019

2019
[22]

Robust dynamic model selection for financial forecasting,

F. Garcia and M. C. Medeiros, “Robust dynamic model selection for financial forecasting,” Journal of Econometrics , vol. 237, no. 2, p. 105526, 2023

2023
[23]

What uncertainties do we need in Bayesian deep learning for computer vision?

A. Kendall and Y . Gal, “What uncertainties do we need in Bayesian deep learning for computer vision?” in Advances in neural information processing systems 30 (NIPS 2017) , 2017, pp. 5574–5584

2017
[24]

Simple and scalable predictive uncertainty estimation using deep ensembles,

B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and scalable predictive uncertainty estimation using deep ensembles,” in Advances in neural information processing systems 30 (NIPS 2017), 2017, pp. 6402– 6413

2017
[25]

Practical Bayesian optimiza- tion of machine learning algorithms,

J. Snoek, H. Larochelle, and R. P. Adams, “Practical Bayesian optimiza- tion of machine learning algorithms,” in Advances in neural information processing systems 25 (NIPS 2012) , 2012, pp. 2951–2959

2012
[26]

Improved training of Wasserstein GANs,

I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. C. Courville, “Improved training of Wasserstein GANs,” in Advances in neural information processing systems 30 (NIPS 2017) , 2017, pp. 5767–5777

2017

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Deep learning for finance: evidence from algorithmic trading in the cryptocurrency market,

J. B. Heaton, N. G. Polson, and J. H. Witte, “Deep learning for finance: evidence from algorithmic trading in the cryptocurrency market,” Ap- plied Stochastic Models in Business and Industry , vol. 33, no. 1, pp. 3–12, 2017

work page 2017

[2] [2]

Deep learning with long short-term memory networks for financial market predictions,

T. Fischer and C. Krauss, “Deep learning with long short-term memory networks for financial market predictions,” European Journal of Oper- ational Research, vol. 270, no. 2, pp. 654–669, 2018

work page 2018

[3] [3]

Dynamic neural architecture search,

X. Zhang, X. Dai, H. Wang, Y . Chen, D. Lin, Y . Qiao, J. Gao, and Y . He, “Dynamic neural architecture search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 23 599–23 609

work page 2023

[4] [4]

A regime- switching model of long-term stock market returns,

P. Nystrup, B. V . Hansen, H. Madsen, and E. Lindstr ¨om, “A regime- switching model of long-term stock market returns,” The European Journal of Finance , vol. 21, no. 13-14, pp. 1155–1175, 2015

work page 2015

[5] [5]

Attention is all you need,

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in neural information processing systems 30 (NIPS 2017) , 2017, pp. 5998–6008

work page 2017

[6] [6]

KAN: Kolmogorov-Arnold Networks,

Z. Liu, Y . Wang, S. Vaidya, F. Ruehle, J. Halatek, Y . Du, J. Schnei- der, T. Yamamoto, S. Nakatsuji, C. Ho, and M. Tegmark, “KAN: Kolmogorov-Arnold Networks,” arXiv preprint arXiv:2404.19756, 2024

arXiv 2024

[7] [7]

N-BEATS: Neural basis expansion analysis for interpretable time series forecasting,

B. N. Oreshkin, D. Carpov, N. Chapados, and Y . Bengio, “N-BEATS: Neural basis expansion analysis for interpretable time series forecasting,” in International Conference on Learning Representations (ICLR) , 2020

work page 2020

[8] [8]

Neural architecture search with Bayesian optimisation and optimal transport,

K. Kandasamy, W. Neiswanger, J. Schneider, B. P ´oczos, and E. Xing, “Neural architecture search with Bayesian optimisation and optimal transport,” in Advances in Neural Information Processing Systems 31 (NeurIPS 2018), 2018, pp. 2016–2025

work page 2018