{"total":12,"items":[{"citing_arxiv_id":"2605.10330","ref_index":28,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Fast Training of Mixture-of-Experts for Time Series Forecasting via Expert Loss Integration","primary_cat":"stat.ML","submitted_at":"2026-05-11T10:33:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A new MoE training method integrates expert-level losses and partial online updates to improve forecasting accuracy and efficiency over standard statistical and neural models.","context_count":1,"top_context_role":"baseline","top_context_polarity":"baseline","context_text":"Exponential Smoothing (SES), Theta [2], Trigonometric Box-Cox ARMA Trend Seasonal [TBATS, 22], Exponential Smoothing [ETS, 12], and Auto-Regressive Integrated Moving Average [ARIMA, 6]. The remaining models are global forecasting approaches, including a linear pooled regression model [PR, 33], CatBoost [29], a feed-forward neural network [FFNN, 10], DeepAR [30], N-BEATS [28], WaveNet [5], and a Transformer model [35]. The forecast comparison between the proposed model and the benchmark models is based on two popular error metrics: Mean Absolute Error [MAE, 1] and Root Mean Squared Error (RMSE). In addition, Mean Absolute Scaled Error [MASE, 15] is considered, as it provides scale-independent evaluation and accounts for seasonality when a seasonal naïve benchmark is used, as is the case"},{"citing_arxiv_id":"2605.09208","ref_index":41,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"TSNN: A Non-parametric and Interpretable Framework for Traffic Time Series Forecasting","primary_cat":"cs.LG","submitted_at":"2026-05-09T22:55:52+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"TSNN matches time series entries to a training-derived memory bank to forecast traffic without any trainable parameters and achieves competitive accuracy on four real-world datasets.","context_count":1,"top_context_role":"baseline","top_context_polarity":"baseline","context_text":"Graph WaveNet [13] proposes a self-learned hidden graph in the graph convolution module and adapts dilated convolution to aggregate temporal dependency. The following studies use a dynamic graph to complement dynamic dependency into the predefined graph [39], [40]. Besides the above models, some studies leveraged mul- tilayer perceptrons (MLPs) and achieved impressive perfor- mance. N-beats [41] provides interpretable outputs by utilizing MLPs and seasonality functions. STID [18] achieves compet- itive performance with only MLPs and spatial and temporal embeddings. Even in the field of long-term forecasting, many models based on linear layers, such as DLinear [42], RLinear [43], challenge the necessity of transformer-based models. More recent studies adapt the Mixture-of-Experts techniques"},{"citing_arxiv_id":"2605.08653","ref_index":21,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"C2L-Net: A Data-Driven Model for State-of-Charge Estimation of Lithium-Ion Batteries During Discharge","primary_cat":"cs.AI","submitted_at":"2026-05-09T03:47:08+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"C2L-Net delivers competitive SOC estimation accuracy on drive-cycle data with up to 60x faster inference by using chunk-based attention, Fourier seasonality, causal GRU encoding, and a recursive-style latest decoder.","context_count":1,"top_context_role":"method","top_context_polarity":"use_method","context_text":"time step is then calculated as e(n) τ =W ah(n) τ +b a,(8) wheree (n) τ ∈R. The normalized attention weight is computed by α(n) τ = exp \u0010 e(n) τ \u0011 PLc j=1 exp \u0010 e(n) j \u0011 .(9) The chunk-level feature representation is obtained using weighted pooling: z(n) = LcX τ=1 α(n) τ h(n) τ ,(10) wherez (n) ∈R d. 9 Seasonality Basic The Seasonality Basis module, inspired by [21], maps the chunk-level feature represen- tation into a coefficient vector, which is then used to generate a token that captures local periodic and trend-related temporal patterns. Specifically, the chunk-level feature representation is first mapped into a coefficient vector: θ(n) =ϕ \u0010 z(n) \u0011 ,(11) whereϕ(·) denotes a feed-forward network implemented by two linear layers with a"},{"citing_arxiv_id":"2605.03795","ref_index":46,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Graph Convolutional Support Vector Regression for Robust Spatiotemporal Forecasting of Urban Air Pollution","primary_cat":"cs.LG","submitted_at":"2026-05-05T14:24:49+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"GCSVR combines graph convolutions for spatial station dependencies with SVR for nonlinear temporal patterns, yielding more accurate and stable air pollution forecasts on Delhi and Mumbai datasets than standard benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.20802","ref_index":26,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Machine Learning-Based Characterization of Solar p-Mode Frequency Shifts during Solar Cycle 25","primary_cat":"astro-ph.SR","submitted_at":"2026-04-22T17:33:03+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"Machine learning and time-series methods are applied to characterize solar p-mode frequency shifts for solar cycle 25 as a potential early indicator of solar activity.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.16240","ref_index":33,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"CollideNet: Hierarchical Multi-scale Video Representation Learning with Disentanglement for Time-To-Collision Forecasting","primary_cat":"cs.CV","submitted_at":"2026-04-17T17:00:03+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"CollideNet achieves state-of-the-art time-to-collision forecasting on three public datasets by combining multi-scale spatial aggregation with temporal disentanglement of trend and seasonality in a hierarchical transformer.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.14994","ref_index":25,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Degradation-aware Predictive Energy Management for Fuel Cell-Battery Ship Power System with Data-driven Load Forecasting","primary_cat":"eess.SY","submitted_at":"2026-04-16T13:20:27+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"A degradation-aware predictive controller for hybrid ship power systems reduces hydrogen consumption by up to 5.8% and fuel cell degradation by up to 36.4% versus a filter-based benchmark on real harbor tug data.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2603.22586","ref_index":15,"ref_count":2,"confidence":0.9,"is_internal_anchor":false,"paper_title":"A Foundation Model for Instruction-Conditioned In-Context Time Series Tasks","primary_cat":"cs.LG","submitted_at":"2026-03-23T21:24:41+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"iAmTime is a time-series foundation model that uses instruction-conditioned in-context learning from demonstrations to perform zero-shot adaptation on forecasting, imputation, classification, and related tasks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.25914","ref_index":13,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"ReNF: Rethinking the Design of Neural Long-Term Time Series Forecasters","primary_cat":"cs.LG","submitted_at":"2025-09-30T08:05:59+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ReNF proposes Boosted Direct Output (BDO) and parameter smoothing so a basic temporal MLP outperforms complex state-of-the-art models on long-term time series forecasting benchmarks by implicitly combining forecasts to reduce uncertainty.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.06503","ref_index":57,"ref_count":2,"confidence":0.9,"is_internal_anchor":false,"paper_title":"An AI system to help scientists write expert-level empirical software","primary_cat":"cs.AI","submitted_at":"2025-09-08T10:08:36+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ERA combines LLMs and tree search to produce expert-level empirical software that outperforms top human methods on single-cell analysis leaderboards and CDC COVID-19 forecasts.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2502.00816","ref_index":18,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Sundial: A Family of Highly Capable Time Series Foundation Models","primary_cat":"cs.LG","submitted_at":"2025-02-02T14:52:50+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Sundial uses TimeFlow Loss for native pre-training of Transformers on continuous time series from TimeBench, achieving SOTA point and probabilistic forecasting with millisecond inference.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2407.13278","ref_index":56,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Deep Time Series Models: A Comprehensive Survey and Benchmark","primary_cat":"cs.LG","submitted_at":"2024-07-18T08:31:55+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"This survey and benchmark of deep time series models using the released TSLib library finds that models with specific structures perform well only on distinct analysis tasks.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"a function space, meaning any function in that space can be expressed as a linear combination of these basis functions. In the context of time series analysis, basis expansion is used to reveal complex non-linear temporal relationships by decomposing the time series into a combination of basic variations, which also enhances interpretability. As a representative model, N-BEATS [56] presents hierarchical decomposition to time series by utilizing a fully connected layer to produce expansion coefficients for both backward and forward forecasts. For l-th blocks in the proposed hierarchical architecture, the operation can be as follows: Xl = Xl−1 − ˆXl−1 ˆXl, ˆYl = Block l(Xl), (4) where ˆXl−1 is the backcast results which restrict the block"}],"limit":50,"offset":0}