BERTO: Intent-Driven Network Time Series Forecasting via Natural Language Operator Preferences

Christian Maciocco; Nitin Priyadarshini Shankar; Sheetal Kalyani; Vaibhav Singh

arxiv: 2512.05721 · v2 · pith:KX2UUM4Gnew · submitted 2025-12-05 · 💻 cs.LG

BERTO: Intent-Driven Network Time Series Forecasting via Natural Language Operator Preferences

Nitin Priyadarshini Shankar , Vaibhav Singh , Sheetal Kalyani , Christian Maciocco This is my paper

Pith reviewed 2026-05-21 17:53 UTC · model grok-4.3

classification 💻 cs.LG

keywords cellular traffic forecastingintent-driven predictionbalancing loss functionprompt conditioningenergy optimizationSLA violationsBERT-based modelsnetwork time series

0 comments

The pith

A single fine-tuned model uses natural language prompts and a balancing loss to shift its traffic forecasts toward under- or over-prediction as operators prefer.

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

The paper shows that standard forecasting models minimize symmetric errors and stay indifferent to changing operational goals such as saving power versus protecting service quality. BERTO combines prompt conditioning with a balancing loss so that one model, after a single fine-tuning pass, can produce forecasts biased in the direction an operator states in plain language. This matters for cellular networks because it removes the need to maintain or retrain separate models when priorities shift. Real-world experiments confirm the same model can span roughly 1.4 kW of power use while tolerating ninefold changes in SLA violations.

Core claim

BERTO is a BERT-based framework for cellular traffic prediction and energy optimization. It achieves high prediction accuracy while letting a single fine-tuned model operate across multiple forecasting regimes via natural-language operator prompts. By pairing a Balancing Loss Function with prompt-based conditioning, the model adaptively shifts its forecasting bias toward underprediction or overprediction according to the operator's chosen trade-off between power savings and service quality, thereby generating different decision-aware forecasts without retraining or parameter changes.

What carries the argument

Prompt-based conditioning combined with the Balancing Loss Function (BLF), which adjusts the model's prediction bias to match operator preferences expressed in natural language.

If this is right

The same model can generate distinct decision-aware forecasts for different operator priorities without retraining.
Forecasting operation spans an approximately 1.4 kW range in power consumption.
The approach accommodates up to ninefold variation in SLA violations.
The resulting flexibility suits intelligent RAN deployments where priorities change over time.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Operators could change forecasting behavior in response to live conditions simply by issuing new language instructions rather than retuning parameters.
The method may reduce the number of separate models that must be stored and maintained for varying network objectives.
Similar prompt-plus-balancing-loss designs could be tested in other time-series settings where user intent dictates whether to err high or low.

Load-bearing premise

Natural language prompts together with the balancing loss function let one fine-tuned model reliably change its forecasting bias across regimes without retraining or modifying its parameters.

What would settle it

A controlled test in which different natural-language prompts requesting opposite biases produce identical forecasts or require parameter updates to achieve the requested shift.

Figures

Figures reproduced from arXiv: 2512.05721 by Christian Maciocco, Nitin Priyadarshini Shankar, Sheetal Kalyani, Vaibhav Singh.

**Figure 1.** Figure 1: System model. for efficient series modeling. Using Autoformer, [7] forecasts wireless network traffic in short time intervals and uses the forecast to dynamically orchestrate and deploy either an RLdriven traffic steering xApp or a cell sleeping rApp. [8] proposes GLSTTN, which combines transformer modules and densely connected CNNs to achieve state-of-the-art accuracy in city-level cellular traffic predi… view at source ↗

**Figure 2.** Figure 2: Solution Model. cells with higher center frequencies or licensed spectrum in scenarios with reduced traffic. The priority order of cells, denoted as C, is assumed to be known or learned over time. The sleep state configuration for the cluster, represented by the vector z, must satisfy the constraints z ∈ C. The future load vector ˆx for the cells is used to optimize power consumption P(x, z) while minimizi… view at source ↗

**Figure 3.** Figure 3: Prompt generation. C. Balancing Loss Function The Balancing Loss Function (BLF) [11] is an asymmetrical loss function designed to balance underprediction and overprediction in machine learning models. It introduces a tunable parameter, q, that controls the penalization asymmetry. Mathematically, BLF is defined as: BLF = max q · (y − yˆ) q + 1 , (ˆy − y) q + 1 , (4) where y − yˆ represents the predicti… view at source ↗

**Figure 4.** Figure 4: BERTO Architecture. uninterrupted service to users. This scheme relies on accurate load predictions and a well-defined threshold to determine the optimal switching points, thus minimizing energy usage while maintaining quality of service. F. Performance Metrics 1) Power model and savings calculation: The power consumption model evaluates the energy usage of cellular networks in both active and inactive s… view at source ↗

read the original abstract

Traditional cellular traffic forecasting models are optimized for minimizing symmetric errors, leaving them indifferent to shifting operational priorities. To bridge this gap, we introduce BERTO, a BERT-based framework for traffic prediction and energy optimization in cellular networks. Built on transformer architectures, BERTO achieves high prediction accuracy while enabling a single fine-tuned model to operate across multiple forecasting regimes via natural-language operator prompts. By combining a Balancing Loss Function (BLF) with prompt-based conditioning, BERTO adaptively shifts its forecasting bias toward underprediction or overprediction depending on the operator's desired trade-off between power savings and service quality. This allows the same model to dynamically generate different decision-aware forecasts without retraining or modifying model parameters. Experiments on real-world datasets demonstrate that BERTO can operate across a flexible range of approximately 1.4 kW in power consumption while balancing 9x variation in service level agreement (SLA) violations, making it well suited for intelligent RAN deployments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

BERTO shows how natural language prompts plus a balancing loss can steer a single BERT forecast model toward power-saving or quality-focused regimes in cellular networks, but the abstract leaves the experimental claims hard to evaluate.

read the letter

The main takeaway is that this paper gives operators a way to bias network traffic forecasts using plain language instructions, so the same model can favor underprediction for energy savings or overprediction for fewer SLA violations without retraining. It combines a BERT backbone with prompt conditioning and a Balancing Loss Function to make that shift happen adaptively. The reported results from real-world data claim coverage of about 1.4 kW in power consumption and 9x variation in SLA violations, which lines up with practical needs in RAN optimization where priorities shift over time. What the work does reasonably well is identify the limitation of symmetric error minimization in standard forecasting and propose a controllable alternative tied to operator intent. The natural language interface is a straightforward addition that could make the system more usable for non-ML teams in telecom. The approach builds on existing transformer time-series work but applies it specifically to decision-aware network forecasting with the added loss term for bias control. The soft spots sit mostly in the experimental reporting. The abstract gives headline performance ranges but skips baselines, precise metrics, validation splits, error analysis, or checks on how consistently the prompt-driven shifts hold across different conditions. That leaves the central claim about reliable multi-regime operation resting on outcomes that are not yet visible in detail. The BLF weighting factors also look like they could require tuning, which might limit how plug-and-play the method really is. This paper is aimed at researchers working on AI for cellular networks or controllable time-series models. Someone focused on intent-based systems or RAN energy management would get the most out of the framing and mechanism. I would send it for peer review. The motivation and high-level design are coherent enough that referees could usefully press for stronger validation and comparisons.

Referee Report

1 major / 0 minor

Summary. The manuscript introduces BERTO, a BERT-based framework for traffic prediction and energy optimization in cellular networks. It claims that by combining a Balancing Loss Function (BLF) with prompt-based conditioning using natural language operator preferences, a single fine-tuned model can adaptively shift its forecasting bias to trade off between power savings and service quality, operating across approximately 1.4 kW in power consumption while balancing 9x variation in SLA violations, without retraining or modifying parameters.

Significance. If the empirical claims hold, the work could enable more flexible decision-aware forecasting in RAN deployments by aligning model outputs with operator intent via natural language without maintaining multiple specialized models. The core idea of prompt-conditioned bias shifting via BLF is a potentially useful contribution to intent-driven time series modeling if supported by rigorous validation.

major comments (1)

Abstract and experimental results: The abstract reports concrete operating ranges (approximately 1.4 kW power consumption and 9x variation in SLA violations) from real-world dataset experiments, yet provides no information on baselines, metrics, validation procedures, or error analysis. This absence directly undermines assessment of whether the BLF plus prompt conditioning reliably achieves the claimed bias shifting across regimes without retraining.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comments and the recommendation for major revision. We address the concern regarding the abstract and experimental presentation below, and we will incorporate changes to improve transparency while preserving the integrity of our results.

read point-by-point responses

Referee: [—] Abstract and experimental results: The abstract reports concrete operating ranges (approximately 1.4 kW power consumption and 9x variation in SLA violations) from real-world dataset experiments, yet provides no information on baselines, metrics, validation procedures, or error analysis. This absence directly undermines assessment of whether the BLF plus prompt conditioning reliably achieves the claimed bias shifting across regimes without retraining.

Authors: We agree that the abstract would benefit from additional context to allow readers to better evaluate the reported operating ranges. The full manuscript details the experimental setup, including comparisons to standard baselines such as LSTM and vanilla Transformer models, evaluation using metrics like mean absolute error alongside power consumption and SLA violation counts, and validation procedures involving real-world cellular traffic datasets with temporal cross-validation and error analysis across multiple regimes. To address the referee's point directly, we will revise the abstract in the next version to briefly reference these elements (baselines, key metrics, and validation approach) without expanding its length excessively. This change will make the claims more self-contained while maintaining focus on the core contribution of prompt-conditioned bias shifting via the Balancing Loss Function. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper introduces BERTO as a new framework combining a Balancing Loss Function (BLF) with natural-language prompt conditioning on a BERT-based transformer. All performance claims (1.4 kW power range, 9× SLA variation) are presented strictly as outcomes of experiments on real-world datasets rather than as quantities derived by construction from fitted parameters or prior self-citations. No equations or logical steps in the provided text reduce a claimed prediction or first-principles result to its own inputs. The central mechanism (prompt-driven bias shift without retraining) is an empirical capability demonstrated by the model, not a definitional identity. This is a standard non-circular empirical ML paper.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the effectiveness of prompt conditioning and the Balancing Loss Function; specific implementation details, parameter counts, and dataset characteristics are not provided in the abstract.

free parameters (1)

BLF weighting factors
The Balancing Loss Function is described as controlling bias shift but its internal weighting or scaling parameters are not specified.

axioms (1)

domain assumption Transformer models such as BERT can be fine-tuned for time-series forecasting in network traffic data
The framework is built directly on transformer architectures for this task.

pith-pipeline@v0.9.0 · 5706 in / 1283 out tokens · 120823 ms · 2026-05-21T17:53:56.741885+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

By combining a Balancing Loss Function (BLF) with prompt-based conditioning, BERTO adaptively shifts its forecasting bias... BLF = max{ q·(y−ŷ)/(q+1), (ŷ−y)/(q+1) }
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

BERT-based architecture fine-tuned specifically for short-term network time series forecasting... Time Series Prediction (TSP) head

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages · 2 internal anchors

[1]

Spatial- temporal cellular traffic prediction for 5g and beyond: A graph neural networks-based approach,

Z. Wang, J. Hu, G. Min, Z. Zhao, Z. Chang, and Z. Wang, “Spatial- temporal cellular traffic prediction for 5g and beyond: A graph neural networks-based approach,”IEEE Transactions on Industrial Informatics, vol. 19, no. 4, pp. 5722–5731, 2023

work page 2023
[2]

Wireless traffic modeling and prediction using seasonal arima models,

Y . Shu, M. Yu, J. Liu, and O. Yang, “Wireless traffic modeling and prediction using seasonal arima models,” inIEEE International Conference on Communications, 2003. ICC ’03., vol. 3, 2003, pp. 1675– 1679 vol.3

work page 2003
[3]

Traffic prediction for mobile network using holt-winter’s exponential smoothing,

D. Tikunov and T. Nishimura, “Traffic prediction for mobile network using holt-winter’s exponential smoothing,” in2007 15th International Conference on Software, Telecommunications and Computer Networks, 2007, pp. 1–5

work page 2007
[4]

Traffic forecasting in cellular networks using the lstm rnn,

A. Dalgkitsis, M. Louta, and G. T. Karetsos, “Traffic forecasting in cellular networks using the lstm rnn,” inProceedings of the 22nd Pan-Hellenic Conference on Informatics, ser. PCI ’18. New York, NY , USA: Association for Computing Machinery, 2018, p. 28–33. [Online]. Available: https://doi.org/10.1145/3291533.3291540

work page doi:10.1145/3291533.3291540 2018
[5]

Informer: Beyond efficient transformer for long sequence time-series forecasting,

H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond efficient transformer for long sequence time-series forecasting,” inProceedings of the AAAI conference on artificial intel- ligence, vol. 35, no. 12, 2021, pp. 11 106–11 115

work page 2021
[6]

Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,

H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,” inAdvances in Neural Information Processing Systems, M. Ranzato, A. Beygelzimer, Y . Dauphin, P. Liang, and J. W. Vaughan, Eds., vol. 34. Curran Associates, Inc., 2021, pp. 22 419–22 430

work page 2021
[7]

Transformer-based wireless traffic prediction and network optimization in o-ran,

M. A. Habib, P. E. I. Rivera, Y . Ozcan, M. Elsayed, M. Bavand, R. Gaigalas, and M. Erol-Kantarci, “Transformer-based wireless traffic prediction and network optimization in o-ran,” in2024 IEEE Inter- national Conference on Communications Workshops (ICC Workshops), 2024, pp. 1–6

work page 2024
[8]

A spatial- temporal transformer network for city-level cellular traffic analysis and prediction,

B. Gu, J. Zhan, S. Gong, W. Liu, Z. Su, and M. Guizani, “A spatial- temporal transformer network for city-level cellular traffic analysis and prediction,”IEEE Transactions on Wireless Communications, vol. 22, no. 12, pp. 9412–9423, 2023

work page 2023
[9]

Large language models for wireless cellular traffic prediction: A multi-timespan approach,

M. H. Shokouhi and V . W. S. Wong, “Large language models for wireless cellular traffic prediction: A multi-timespan approach,” inGLOBECOM 2024 - 2024 IEEE Global Communications Conference, 2024, pp. 1293– 1298

work page 2024
[10]

Spectrum- llm: Large language models for next-generation spectrum prediction,

C. Liu, Y . Wang, S. Mao, D. Niyato, X. Wang, and G. Gui, “Spectrum- llm: Large language models for next-generation spectrum prediction,” IEEE Wireless Communications, pp. 1–7, 2025

work page 2025
[11]

Intelligent ran power saving using balanced model training in cellular networks,

V . Singh, M. Gupta, and C. Maciocco, “Intelligent ran power saving using balanced model training in cellular networks,” in2022 20th International Symposium on Modeling and Optimization in Mobile, Ad hoc, and Wireless Networks (WiOpt), 2022, pp. 357–364

work page 2022
[12]

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,”Advances in neural information processing systems, vol. 30, 2017

work page 2017
[13]

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

J. Devlin, “BERT: Pre-training of deep bidirectional transformers for language understanding,”arXiv preprint arXiv:1810.04805, 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[14]

Inter- ference prediction in partially loaded cellular networks using asymmetric cost functions,

S. Parthasarathy, S. K. Pulliyakode, S. Kalyani, and R. K. Ganti, “Inter- ference prediction in partially loaded cellular networks using asymmetric cost functions,”IEEE Communications Letters, vol. 22, no. 6, pp. 1288– 1291, 2018

work page 2018
[15]

Load balancing in o-ran,

H. Zafar, E. Tohidi, M. Kasparick, and S. Sta ´nczak, “Load balancing in o-ran,” in2024 IEEE Wireless Communications and Networking Conference (WCNC), 2024, pp. 1–6

work page 2024
[16]

A flexible and future-proof power model for cellular base stations,

B. Debaillie, C. Desset, and F. Louagie, “A flexible and future-proof power model for cellular base stations,” in2015 IEEE 81st Vehicular Technology Conference (VTC Spring), 2015, pp. 1–7

work page 2015
[17]

Feed-forward neural networks,

G. Bebis and M. Georgiopoulos, “Feed-forward neural networks,”IEEE Potentials, vol. 13, no. 4, pp. 27–31, 1994

work page 1994
[18]

Chronos: Learning the Language of Time Series

A. F. Ansari, L. Stella, C. Turkmen, X. Zhang, P. Mercado, H. Shen, O. Shchur, S. S. Rangapuram, S. P. Arango, S. Kapooret al., “Chronos: Learning the language of time series,”arXiv preprint arXiv:2403.07815, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024

[1] [1]

Spatial- temporal cellular traffic prediction for 5g and beyond: A graph neural networks-based approach,

Z. Wang, J. Hu, G. Min, Z. Zhao, Z. Chang, and Z. Wang, “Spatial- temporal cellular traffic prediction for 5g and beyond: A graph neural networks-based approach,”IEEE Transactions on Industrial Informatics, vol. 19, no. 4, pp. 5722–5731, 2023

work page 2023

[2] [2]

Wireless traffic modeling and prediction using seasonal arima models,

Y . Shu, M. Yu, J. Liu, and O. Yang, “Wireless traffic modeling and prediction using seasonal arima models,” inIEEE International Conference on Communications, 2003. ICC ’03., vol. 3, 2003, pp. 1675– 1679 vol.3

work page 2003

[3] [3]

Traffic prediction for mobile network using holt-winter’s exponential smoothing,

D. Tikunov and T. Nishimura, “Traffic prediction for mobile network using holt-winter’s exponential smoothing,” in2007 15th International Conference on Software, Telecommunications and Computer Networks, 2007, pp. 1–5

work page 2007

[4] [4]

Traffic forecasting in cellular networks using the lstm rnn,

A. Dalgkitsis, M. Louta, and G. T. Karetsos, “Traffic forecasting in cellular networks using the lstm rnn,” inProceedings of the 22nd Pan-Hellenic Conference on Informatics, ser. PCI ’18. New York, NY , USA: Association for Computing Machinery, 2018, p. 28–33. [Online]. Available: https://doi.org/10.1145/3291533.3291540

work page doi:10.1145/3291533.3291540 2018

[5] [5]

Informer: Beyond efficient transformer for long sequence time-series forecasting,

H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond efficient transformer for long sequence time-series forecasting,” inProceedings of the AAAI conference on artificial intel- ligence, vol. 35, no. 12, 2021, pp. 11 106–11 115

work page 2021

[6] [6]

Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,

H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,” inAdvances in Neural Information Processing Systems, M. Ranzato, A. Beygelzimer, Y . Dauphin, P. Liang, and J. W. Vaughan, Eds., vol. 34. Curran Associates, Inc., 2021, pp. 22 419–22 430

work page 2021

[7] [7]

Transformer-based wireless traffic prediction and network optimization in o-ran,

M. A. Habib, P. E. I. Rivera, Y . Ozcan, M. Elsayed, M. Bavand, R. Gaigalas, and M. Erol-Kantarci, “Transformer-based wireless traffic prediction and network optimization in o-ran,” in2024 IEEE Inter- national Conference on Communications Workshops (ICC Workshops), 2024, pp. 1–6

work page 2024

[8] [8]

A spatial- temporal transformer network for city-level cellular traffic analysis and prediction,

B. Gu, J. Zhan, S. Gong, W. Liu, Z. Su, and M. Guizani, “A spatial- temporal transformer network for city-level cellular traffic analysis and prediction,”IEEE Transactions on Wireless Communications, vol. 22, no. 12, pp. 9412–9423, 2023

work page 2023

[9] [9]

Large language models for wireless cellular traffic prediction: A multi-timespan approach,

M. H. Shokouhi and V . W. S. Wong, “Large language models for wireless cellular traffic prediction: A multi-timespan approach,” inGLOBECOM 2024 - 2024 IEEE Global Communications Conference, 2024, pp. 1293– 1298

work page 2024

[10] [10]

Spectrum- llm: Large language models for next-generation spectrum prediction,

C. Liu, Y . Wang, S. Mao, D. Niyato, X. Wang, and G. Gui, “Spectrum- llm: Large language models for next-generation spectrum prediction,” IEEE Wireless Communications, pp. 1–7, 2025

work page 2025

[11] [11]

Intelligent ran power saving using balanced model training in cellular networks,

V . Singh, M. Gupta, and C. Maciocco, “Intelligent ran power saving using balanced model training in cellular networks,” in2022 20th International Symposium on Modeling and Optimization in Mobile, Ad hoc, and Wireless Networks (WiOpt), 2022, pp. 357–364

work page 2022

[12] [12]

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,”Advances in neural information processing systems, vol. 30, 2017

work page 2017

[13] [13]

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

J. Devlin, “BERT: Pre-training of deep bidirectional transformers for language understanding,”arXiv preprint arXiv:1810.04805, 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018

[14] [14]

Inter- ference prediction in partially loaded cellular networks using asymmetric cost functions,

S. Parthasarathy, S. K. Pulliyakode, S. Kalyani, and R. K. Ganti, “Inter- ference prediction in partially loaded cellular networks using asymmetric cost functions,”IEEE Communications Letters, vol. 22, no. 6, pp. 1288– 1291, 2018

work page 2018

[15] [15]

Load balancing in o-ran,

H. Zafar, E. Tohidi, M. Kasparick, and S. Sta ´nczak, “Load balancing in o-ran,” in2024 IEEE Wireless Communications and Networking Conference (WCNC), 2024, pp. 1–6

work page 2024

[16] [16]

A flexible and future-proof power model for cellular base stations,

B. Debaillie, C. Desset, and F. Louagie, “A flexible and future-proof power model for cellular base stations,” in2015 IEEE 81st Vehicular Technology Conference (VTC Spring), 2015, pp. 1–7

work page 2015

[17] [17]

Feed-forward neural networks,

G. Bebis and M. Georgiopoulos, “Feed-forward neural networks,”IEEE Potentials, vol. 13, no. 4, pp. 27–31, 1994

work page 1994

[18] [18]

Chronos: Learning the Language of Time Series

A. F. Ansari, L. Stella, C. Turkmen, X. Zhang, P. Mercado, H. Shen, O. Shchur, S. S. Rangapuram, S. P. Arango, S. Kapooret al., “Chronos: Learning the language of time series,”arXiv preprint arXiv:2403.07815, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024