Pith. sign in

REVIEW 4 major objections 5 minor 9 cited by

Adaptive Cache Management for Complex Storage Systems Using CNN-LSTM-Based Spatiotemporal Prediction

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a CNN-LSTM hybrid predicts storage cache demand more accurately than LRU, LFU, RNN, GRU-RNN, and plain LSTM, and that the better forecasts can drive cache decisions.

desk verdict Thin CNN-LSTM cache paper: the central hit-rate claim is never measured, and the reported MSE/MAE could be in-sample training error. read the letter →

arxiv 2411.12161 v1 pith:TH2TB3VJ submitted 2024-11-19 cs.DC

Xiaoye Wang , Xuan Li , Linji Wang , Tingyi Ruan , Pochun Li This is my paper · ORCID
classification cs.DC
keywords cachemanagementCNN-LSTMdemandpredictionhitratestoragesystemoptimizationdeeplearningspatiotemporalaccesspatternanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that cache management in storage systems can be treated as a forecasting problem: learn from historical access logs to predict future cache demand, then use those predictions to decide which data blocks to prefetch and which to demote. It proposes a CNN-LSTM model that combines convolutional feature extraction with long short-term memory to capture the spatial and temporal structure of access patterns. On a real storage-system trace dataset, the model reports lower mean-squared and mean-absolute prediction error than LRU, LFU, RNN, GRU-RNN, and plain LSTM, with the largest gains over the rule-based policies. The paper takes these prediction-error improvements as evidence that the model would raise cache hit rates and system performance.

What carries the argument

The mechanism is a two-stage neural architecture trained on a time-series feature matrix $X \in \mathbb{R}^{T \times N}$ of historical access statistics. A convolutional layer produces a refined representation $X' = f(W * X + b)$, where $W$ is the convolution kernel, $b$ a bias, and $f$ an activation function; this is meant to capture spatial patterns such as access frequency and preference distribution. The refined features are fed into an LSTM whose input, forget, and output gates update a memory cell $c_t$ and hidden state $h_t$, capturing long-term dependencies in access behavior. A fully connected layer maps the final hidden state to a predicted cache demand $\hat{y}_t$, and the network is trained by minimizing mean squared error $L = \frac{1}{T}\sum_{t}(y_t - \hat{y}_t)^2$. This prediction, rather than a fixed rule, is the object that carries the proposed cache-management strategy.

What would settle it

Replay the same real storage traces through a cache whose prefetch and eviction decisions follow CNN-LSTM predictions, and compare the resulting hit rate and latency against LRU and LFU; if the lower MSE and MAE do not produce a better hit rate, the paper's practical claim is false. A second check is to show whether the predicted scalar $\hat{y}_t$ corresponds to any block-level reuse signal; if it does not, the forecast cannot guide cache decisions.

Watch

Extended reading notes

Core claim

The central discovery, on the paper's own terms, is that a CNN-LSTM hybrid predicts future cache demand better than both rule-based cache policies and simpler recurrent models. The authors report MSE and MAE values of 0.244 and 0.127 for CNN-LSTM, versus 0.375 and 0.321 for LSTM, 0.521 and 0.465 for GRU-RNN, 0.623 and 0.546 for RNN, 0.873 and 0.725 for LFU, and 0.951 and 0.867 for LRU on the same real storage traces. They attribute the gap to the combination of CNN's ability to extract spatial access features, such as frequency and preference distribution, and LSTM's ability to model long-term temporal dependencies. The conclusion drawn is that these forecasts can drive dynamic cache allocation and thereby improve hit rate and system responsiveness.

Load-bearing premise

The load-bearing premise is that the model's predicted 'cache demand' is a quantity whose accuracy actually governs which blocks the cache keeps; the paper never measures hit rate, latency, or throughput to confirm this.

Editorial extensions

If this is right

  • Cache policies could shift from recency and frequency heuristics to learned forecasts, preloading predicted-hot blocks and demoting predicted-cold blocks.
  • The consistent error reduction from rule-based to deep-learning models indicates that complex access patterns contain predictive structure that fixed rules miss.
  • Because the model can keep updating its parameters as new access logs arrive, the cache strategy can adapt to workload drift without manual reconfiguration.
  • If the forecast-to-cache mapping works as described, storage systems should see fewer misses, lower latency, and higher throughput under changing workloads.

Reading between the lines

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

  • The paper reports only aggregate MSE and MAE; the decisive test of its practical claim is a replay experiment on the same traces that measures actual hit rate under CNN-LSTM-driven cache decisions versus LRU and LFU.
  • The advantage of CNN-LSTM over plain LSTM is likely workload-dependent: on access patterns with little spatial structure, the convolutional stage may add little and the model should degrade toward LSTM-like performance.
  • A concrete extension would define the predicted target $\hat{y}_t$ at block or object granularity and evaluate top-$k$ prefetch precision, which would connect prediction error directly to cache hit rate.
  • Reporting variance across multiple training runs would show whether the reported gaps are stable or an artifact of a single 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.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes a CNN-LSTM model for cache demand prediction in storage systems. The method uses CNN to extract spatial features from a multi-dimensional access-history matrix and LSTM to model temporal dependencies, then maps hidden states to a scalar cache-demand value through a fully connected layer. Experiments use the Microsoft Research Traces dataset and report MSE and MAE for LRU, LFU, RNN, GRU-RNN, LSTM, and the proposed CNN-LSTM, which achieves the lowest errors. The conclusion claims that this prediction accuracy translates into improved cache hit rate and system performance. The paper also acknowledges a limitation concerning hardware dependence.

Significance. If the claim were substantiated, a CNN-LSTM predictor with lower demand-prediction error than RNN/GRU/LSTM baselines could be a useful component for learned cache management, particularly given the use of a real-world trace dataset and the inclusion of both traditional and deep-learning comparators. However, the significance is currently limited by the absence of any cache simulation or system-level measurement: the paper reports only prediction errors, never the hit-rate, latency, or throughput effects that are the stated motivation. The manuscript also lacks the experimental protocol (data split, hyperparameters, repeated runs) needed to establish that the reported errors are predictive rather than fitted.

major comments (4)
  1. [Section III.B, Table 1] The central claim in Section IV that the CNN-LSTM model 'effectively improves cache hit rate and system performance' is not supported by the paper's only quantitative evidence: Table 1 reports MSE and MAE but no cache hit rate, access latency, or throughput. Section II states that predictions will be used to load data blocks in advance and migrate low-access blocks, but no experiment evaluates this procedure; hence the load-bearing practical claim is asserted, not tested.
  2. [Section II, prediction target] The target variable y_t is never concretely defined. The text says y_t is 'the predicted value of cache demand at time t', but it does not specify whether this is a per-block access count, a total request count, a byte count, or something else, nor how the feature matrix X (with T time steps and N features) is constructed from the Microsoft trace fields. Without this definition, the MSE and MAE values in Table 1 are not interpretable, and the comparison across models is not meaningful.
  3. [Section III.B, Figure 2] The paper reports no train/validation/test split and Figure 2 shows only the training loss. Since the MSE loss in Section II is the same metric reported in Table 1, the reported errors may describe in-sample training fit rather than held-out prediction. The absence of error bars, repeated runs, or statistical tests means the claimed superiority of CNN-LSTM over LSTM (MSE 0.244 vs. 0.375) cannot be distinguished from random variation.
  4. [Section III.B, comparators] LRU and LFU are cache replacement policies, not demand predictors. Computing MSE/MAE for them requires defining an explicit prediction of y_t, which the paper does not provide. Comparing prediction errors of rule-based replacement policies against neural predictors is therefore mismatched and does not support the paper's conclusion that CNN-LSTM is superior for cache demand prediction.
minor comments (5)
  1. [Section III.B] The experimental setup inconsistently calls the fourth baseline 'GRU-CNN combination model' in the text but 'GRU-RNN' in Table 1; these are different architectures.
  2. [Section II] The equations appear garbled in the manuscript (e.g., the LSTM gate equations and the loss function are not typeset correctly), making the method difficult to follow.
  3. [Section II] The notation 'NTRX' for the feature matrix is unclear; it should be R^{T×N} or explicitly defined.
  4. [References] Many references (e.g., [3]–[5], [9]–[15], [20]–[22]) are unrelated to cache management or storage systems and are not used to support the specific technical claims in the text; a focused reference list would strengthen the paper.
  5. [Figure 1] Figure 1 is referenced as 'the overall architecture of the model' but has no explanatory caption or labels describing the data flow, so the reader cannot map it to Section II.

Circularity Check

1 steps flagged · score 6.0 of 10

The claimed prediction superiority and the cache hit-rate gains rest solely on MSE/MAE of an undefined target, where the evaluation metric equals the training loss and no held-out split is reported, so the reported accuracy may be the in-sample fit.

  1. fitted input called prediction [Section II defines the MSE loss; Section III-B reports Table 1 and Figure 2 (the 'training loss drop graph'); Section IV draws the conclusion.]
    "we use Mean Squared Error (MSE) as the loss function ... By minimizing this loss function, we can adjust the parameters of the model to make the predicted value as close to the actual value as possible. ... The experimental results are shown in Table 1 [MSE column]. ... Finally, we also show images of training loss reduction, as shown in Figure 2. ... the CNN-LSTM model proposed in this article achieved the best experimental results, with an MSE of 0.244 and a MAE of 0.127."

    The MSE/MAE in Table 1, offered as 'evaluation indicators to measure the prediction accuracy,' is the same function as the training loss L=(1/N)Σ(y_t−y'_t)² that Section II minimizes ('we use Mean Squared Error (MSE) as the loss function... to make the predicted value as close to the actual value as possible'). No train/test/validation split is described anywhere, and the only loss evidence is the 'training loss drop graph' (Figure 2). The evidence for 'the CNN-LSTM model performs best in cache demand prediction' is therefore, as far as the text discloses, the in-sample training error — a quantity the optimizer directly minimizes, not a forecast.

full rationale

The paper's central claim — CNN-LSTM 'performs best in cache demand prediction, effectively improving cache hit rate and system performance' — is supported only by the MSE/MAE values of Table 1. That evidence is burdened by circularity: the loss function chosen in Section II is exactly the MSE reported as the evaluation metric, and the paper never states that the table records held-out test error. The only loss curve shown is Figure 2, a 'training loss drop graph,' so the text does not rule out that the reported errors are the in-sample training fit, which the optimizer minimizes by construction rather than a demonstration of predictive skill. This is the fitted-input-called-prediction pattern. I score it 6 rather than higher because a proper train/test split would in principle rescue the MSE comparison as genuine evidence; the paper simply omits any such protocol. No self-citation load-bearing chain exists: reference [19] and the other cited works have no overlap with the present authors, so the self-citation and uniqueness-import patterns do not apply. The undefined prediction target y_t, the comparison of cache policies (LRU/LFU) as if they were predictors with well-defined errors, and the unmeasured hit-rate/latency/throughput claims are correctness risks rather than circular steps, but they compound the central issue: the sole quantitative evidence for the practical conclusion is the quantity the model was trained to minimize, with no reported evaluation protocol that would distinguish fitting from prediction. The limitation passage (hardware dependence) is genuine but does not address this missing held-out evaluation, so it does not reduce the circularity burden.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new theory. It relies on standard CNN/LSTM formulas, on the domain assumption that historical access predicts future demand, and on the proxy assumption that lower MSE/MAE means better cache hit rate. All model hyperparameters, feature definitions, and the prediction target are unreported free choices, so the empirical result is not auditable.

free parameters (3)
  • CNN-LSTM hyperparameters (kernel size, layers, hidden units, learning rate, batch size, sequence length) = not reported
    Section II gives only generic formulas; no values are provided, and the reported results depend entirely on these choices.
  • Input feature matrix dimensions (T, N) and feature encoding = not reported
    Section III-A lists features such as access frequency and block size, but not how they are encoded, normalized, or windowed into the matrix X.
  • Cache-demand target definition y_t = not reported
    The paper never specifies what the scalar cache demand at time t represents, so the fitted target is undefined and the MSE/MAE values are not interpretable.
assumptions (3)
  • domain assumption Historical access patterns contain learnable spatiotemporal structure that predicts future cache demand.
    Section II and Section III assume this without evidence; if access patterns are not predictable, the whole method fails.
  • domain assumption Lower prediction error (MSE/MAE) translates into higher cache hit rate and better system performance.
    Section II states that predictions enable cache prefetching, but no end-to-end experiment measures hit rate, latency, or throughput.
  • standard math Standard CNN and LSTM update equations are correct and applicable as written.
    Section II uses textbook formulas for convolution and LSTM gates, which are accepted background knowledge.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Cache Management for Complex Storage Systems Using CNN-LSTM-Based Spatiotemporal Prediction." pith.science (2026). https://pith.science/paper/TH2TB3VJ

@misc{pith2026241112161,
  author       = {Pith},
  title        = {Pith review of: Adaptive Cache Management for Complex Storage Systems Using CNN-LSTM-Based Spatiotemporal Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TH2TB3VJ}},
  note         = {Machine review of arXiv:2411.12161}
}
read the original abstract

This paper proposes an intelligent cache management strategy based on CNN-LSTM to improve the performance and cache hit rate of storage systems. Through comparative experiments with traditional algorithms (such as LRU and LFU) and other deep learning models (such as RNN, GRU-RNN and LSTM), the results show that the CNN-LSTM model has significant advantages in cache demand prediction. The MSE and MAE values of this model are significantly reduced, proving its effectiveness under complex data access patterns. This study not only verifies the potential of deep learning technology in storage system optimization, but also provides direction and reference for further optimizing and improving cache management strategies. This intelligent cache management strategy performs well in complex storage environments. By combining the spatial feature extraction capabilities of convolutional neural networks and the time series modeling capabilities of long short-term memory networks, the CNN-LSTM model can more accurately predict cache needs, thereby Dynamically optimize cache allocation to improve system response speed and resource utilization. This research provides theoretical support and practical reference for cache optimization under large-scale data access modes, and is of great significance to improving the performance of future storage systems.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Collaborative Optimization in Financial Data Mining Through Deep Learning and ResNeXt

    cs.LG 2024-12 reject novelty 3.0 of 10

    A ResNeXt-based multi-task learning model reportedly outperforms LSTM, Transformer, MCCNN, and DSN on S&P 500 classification and regression, but the experiments lack error bars, code, and leakage controls.

  2. Enhancing Recommendation Systems with GNNs and Addressing Over-Smoothing

    cs.IR 2024-12 reject novelty 3.0 of 10

    Adding initial residual connections and identity mapping to a LightGCN-style recommendation model yields small reported gains on Gowalla, Yelp-2018, and Amazon-Book.

  3. Computer Vision-Driven Gesture Recognition: Toward Natural and Intuitive Human-Computer

    cs.CV 2024-12 reject novelty 2.0 of 10

    A CNN-LSTM gesture recognizer with a decorative 3D skeleton visualization that reports unverifiable accuracy and speed numbers.

  4. Adaptive User Interface Generation Through Reinforcement Learning: A Data-Driven Approach to Personalization and Optimization

    cs.HC 2024-12 reject novelty 2.0 of 10

    A DQN-based reinforcement learning system is reported to reach CTR 0.78 and RR 0.83 on an unverified CLIP Interactions dataset, beating five baselines, but no reproducible evidence is provided.

  5. Dynamic User Interface Generation for Enhanced Human-Computer Interaction Using Variational Autoencoders

    cs.HC 2024-12 reject novelty 2.0 of 10

    A vanilla VAE trained on the RICO dataset is reported to beat AE, GAN, cGAN, DBN, and VAE-GAN on SSIM/MAE for UI generation, but the comparison is under-specified and the dynamic feedback claim is untested.

  6. Optimizing Multi-Task Learning for Enhanced Performance in Large Language Models

    cs.CL 2024-12 reject novelty 2.0 of 10

    A multi-task GPT-4 model is said to beat single-task GPT-4, GPT-3, BERT, and Bi-LSTM on classification and summarization, but the experimental evidence is not reported.

  7. Advanced Risk Prediction and Stability Assessment of Banks Using Time Series Transformer Models

    q-fin.RM 2024-12 reject novelty 2.0 of 10

    A standard Time Series Transformer is compared with five baselines on the UCI Bank Marketing dataset and reported as best for bank stability prediction, but the dataset contains no bank stability index.

  8. An Automated Data Mining Framework Using Autoencoders for Feature Extraction and Dimensionality Reduction

    cs.LG 2024-12 reject novelty 2.0 of 10

    An autoencoder is compared with five dimensionality reduction methods on one UCI dataset and reported to have the best reconstruction error, without error bars or released code.

  9. Leveraging Semi-Supervised Learning to Enhance Data Mining for Image Classification under Limited Labeled Data

    cs.CV 2024-11 reject novelty 1.0 of 10

    A self-training CNN on 10,000 labeled CIFAR-10 images reaches 0.897 accuracy, but missing implementation details and baseline comparisons make the result unverifiable.

Reference graph

Works this paper leans on

23 extracted references · 8 canonical work pages · cited by 9 Pith papers

  1. [23]

    RC-RNN: Reconfigurable Cache Architecture for Storage Systems Using Recurrent Neural Networks,

    S. Ebrahimi, R. Salkhordeh, S. A. Osia, A. Taheri, H. R. Rabiee and H. Asadi, "RC-RNN: Reconfigurable Cache Architecture for Storage Systems Using Recurrent Neural Networks," IEEE Transactions on Emerging Topics in Computing, vol. 10, no. 3, pp. 1492-1506, 2021

    work page 2021

  2. [1]

    An Intelligent Caching Approach in Mobile Edge Computing Environment,

    M. M. Moayeri and H. Momeni, "An Intelligent Caching Approach in Mobile Edge Computing Environment," Proceedings of the 2024 20th CSI International Symposium on Artificial Intelligence and Signal Processing (AISP), IEEE, pp. 1-5, 2024

    work page 2024

  3. [2]

    STRCacheML: A Machine Learning-Assisted Content Caching Policy for Streaming Services,

    A. Mahara, J. Fuentes, C. Poellabauer, et al., "STRCacheML: A Machine Learning-Assisted Content Caching Policy for Streaming Services," CS & IT Conference Proceedings, vol. 14, no. 4, 2024

    work page 2024

  4. [3]

    Predicting Liquidity Coverage Ratio with Gated Recurrent Units: A Deep Learning Model for Risk Management,

    Z. Xu, J. Pan, S. Han, H. Ouyang, Y. Chen and M. Jiang, "Predicting Liquidity Coverage Ratio with Gated Recurrent Units: A Deep Learning Model for Risk Management," arXiv preprint arXiv:2410.19211, 2024

    arXiv 2024

  5. [4]

    Survival prediction across diverse cancer types using neural networks

    X. Yan, W. Wang, M. Xiao, Y. Li, and M. Gao, "Survival prediction across diverse cancer types using neural networks", Proceedings of the 2024 7th International Conference on Machine Vision and Applications, pp. 134-138, 2024

    2024

  6. [5]

    Financial Risk Analysis Using Integrated Data and Transformer-Based Deep Learning,

    Y. Wei, K. Xu, J. Yao, M. Sun and Y. Sun, "Financial Risk Analysis Using Integrated Data and Transformer-Based Deep Learning," Journal of Computer Science and Software Applications, vol. 7, no. 4, pp. 1-8, 2024

    2024

  7. [6]

    Adversarial Neural Networks in Medical Imaging Advancements and Challenges in Semantic Segmentation,

    H. Liu, B. Zhang, Y. Xiang, Y. Hu, A. Shen and Y. Lin, "Adversarial Neural Networks in Medical Imaging Advancements and Challenges in Semantic Segmentation," arXiv preprint arXiv:2410.13099, 2024

    arXiv 2024

  8. [7]

    Deep Learning for Medical Text Processing: BERT Model Fine-Tuning and Comparative Study,

    J. Hu, Y. Cang, G. Liu, M. Wang, W. He and R. Bao, "Deep Learning for Medical Text Processing: BERT Model Fine-Tuning and Comparative Study," arXiv preprint arXiv:2410.20792, 2024

    arXiv 2024

Show all 23 references
  1. [8]

    How to Cache Important Contents for Multi-modal Service in Dynamic Networks: A DRL-based Caching Scheme,

    Z. Zhang, M. St-Hilaire, X. Wei, et al., "How to Cache Important Contents for Multi-modal Service in Dynamic Networks: A DRL-based Caching Scheme," IEEE Transactions on Multimedia, 2024

    2024

  2. [9]

    Efficient and Aesthetic UI Design with a Deep Learning-Based Interface Generation Tree Algorithm,

    S. Duan, R. Zhang, M. Chen, Z. Wang and S. Wang, "Efficient and Aesthetic UI Design with a Deep Learning-Based Interface Generation Tree Algorithm," arXiv preprint arXiv:2410.17586, 2024

    2024 arXiv

  3. [10]

    Research on Intelligent System of Medical Image Recognition and Disease Diagnosis Based on Big Data,

    Y. Zi, X. Cheng, T. Mei, Q. Wang, Z. Gao and H. Yang, "Research on Intelligent System of Medical Image Recognition and Disease Diagnosis Based on Big Data," Proceedings of the 2024 IEEE 2nd International Conference on Image Processing and Computer Applications (ICIPCA), pp. 82...

    2024

  4. [11]

    Deep Learning with HM- VGG: AI Strategies for Multi-modal Image Analysis,

    J. Du, Y. Cang, T. Zhou, J. Hu and W. He, "Deep Learning with HM- VGG: AI Strategies for Multi-modal Image Analysis," arXiv preprint arXiv:2410.24046, 2024

    2024 arXiv

  5. [12]

    A Systematic Study on the Privacy Protection Mechanism of Natural Language Processing in Medical Health Records,

    X. Fei, S. Chai, W. He, L. Dai, R. Xu and L. Cai, "A Systematic Study on the Privacy Protection Mechanism of Natural Language Processing in Medical Health Records," Proceedings of the 2024 IEEE 2nd International Conference on Sensors, Electronics and Computer Engineering (ICSE...

    2024

  6. [13]

    Automated Genre-Aware Article Scoring and Feedback Using Large Language Models,

    C. Wang, Y. Dong, Z. Zhang, R. Wang, S. Wang and J. Chen, "Automated Genre-Aware Article Scoring and Feedback Using Large Language Models," arXiv preprint arXiv:2410.14165, 2024

    2024 arXiv

  7. [14]

    A Recommendation Model Utilizing Separation Embedding and Self- Attention for Feature Mining,

    W. Liu, R. Wang, Y. Luo, J. Wei, Z. Zhao and J. Huang, "A Recommendation Model Utilizing Separation Embedding and Self- Attention for Feature Mining," arXiv preprint arXiv:2410.15026, 2024

    2024 arXiv

  8. [15]

    Optimizing Retrieval-Augmented Generation with Elasticsearch for Enhanced Question-Answering Systems,

    J. Chen, R. Bao, H. Zheng, Z. Qi, J. Wei and J. Hu, "Optimizing Retrieval-Augmented Generation with Elasticsearch for Enhanced Question-Answering Systems," arXiv preprint arXiv:2410.14167, 2024

    2024 arXiv

  9. [16]

    Self- Supervised Graph Neural Networks for Enhanced Feature Extraction in Heterogeneous Information Networks,

    J. Wei, Y. Liu, X. Huang, X. Zhang, W. Liu and X. Yan, "Self- Supervised Graph Neural Networks for Enhanced Feature Extraction in Heterogeneous Information Networks," arXiv preprint arXiv:2410.17617, 2024

    2024 arXiv

  10. [17]

    Advanced RAG Models with Graph Structures: Optimizing Complex Knowledge Reasoning and Text Generation,

    Y. Dong, S. Wang, H. Zheng, J. Chen, Z. Zhang and C. Wang, "Advanced RAG Models with Graph Structures: Optimizing Complex Knowledge Reasoning and Text Generation," arXiv preprint arXiv:2411.03572, 2024

    2024 arXiv

  11. [18]

    Transformers in Opinion Mining: Addressing Semantic Complexity and Model Challenges in NLP,

    J. Du, Y. Jiang and Y. Liang, "Transformers in Opinion Mining: Addressing Semantic Complexity and Model Challenges in NLP," Transactions on Computational and Scientific Methods, vol. 4, no. 10, 2024

    2024

  12. [19]

    Transforming Multidimensional Time Series into Interpretable Event Sequences for Advanced Data Mining

    X. Yan, Y. Jiang, W. Liu, D. Yi, and J. Wei, "Transforming Multidimensional Time Series into Interpretable Event Sequences for Advanced Data Mining", arXiv preprint, arXiv:2409.14327, 2024

    2024 arXiv

  13. [20]

    Leveraging Deep Learning Techniques for Enhanced Analysis of Medical Textual Data,

    Y. Cang, Y. Zhong, R. Ji, Y. Liang, Y. Lei and J. Wang, "Leveraging Deep Learning Techniques for Enhanced Analysis of Medical Textual Data," Proceedings of the 2024 IEEE 2nd International Conference on Sensors, Electronics and Computer Engineering (ICSECE), pp. 1259- 1263, Aug. 2024

    2024

  14. [21]

    Optimizing YOLOv5s Object Detection through Knowledge Distillation Algorithm,

    G. Huang, A. Shen, Y. Hu, J. Du, J. Hu and Y. Liang, "Optimizing YOLOv5s Object Detection through Knowledge Distillation Algorithm," arXiv preprint arXiv:2410.12259, 2024

    2024 arXiv

  15. [22]

    Balancing Innovation and Privacy: Data Security Strategies in Natural Language Processing Applications,

    S. Liu, G. Liu, B. Zhu, Y. Luo, L. Wu and R. Wang, "Balancing Innovation and Privacy: Data Security Strategies in Natural Language Processing Applications," arXiv preprint arXiv:2410.08553, 2024

    2024 arXiv

Pith tools

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