@researai/deepscientist 1.5.16 → 1.6.0

package/AISB/catalog/aisb.t3.087_moses.yaml

@@ -0,0 +1,167 @@
+id: aisb.t3.087_moses
+name: MoSES - Multi-Task Vehicle Routing Solver
+version: 0.1.0
+one_line: Multi-task vehicle-routing solver using LoRA-based specialized experts and
+  adaptive gating under State-Decomposable MDP framework.
+task_description: 'This packaged benchmark covers the Mixture-of-Specialized-Experts
+  Solver (MoSES) for multi-task vehicle routing problems. The approach introduces
+  a State-Decomposable MDP (SDMDP) that reformulates VRPs by expressing the state
+  space as the Cartesian product of basis state spaces, each associated with a basis
+  VRP variant. Basis policies are realized through specialized LoRA experts, each
+  fine-tuned from a frozen pretrained backbone for a specific basis variant. A latent-space
+  extension (LS-SDMDP) incorporates optimal basis policies and a learnable mixture
+  function to enable policy reuse in the embedding space. The mixture function maps
+  basis state embeddings generated by optimal basis policies to unified state embeddings
+  via an adaptive gating mechanism. Under mild assumptions, LS-SDMDP provably recovers
+  the optimal unified policy. The current local snapshot preserves benchmark metadata
+  and paper artifacts but does not contain executable training or evaluation code;
+  restoration of the runnable pipeline is required before benchmark execution.
+
+  '
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 1d+
+cost_band: high
+difficulty: hard
+data_access: public
+primary_outputs:
+- optimality_gap_pct
+- tour_cost
+- inference_time
+launch_profiles:
+- id: analysis_only
+  label: Analysis Only
+  description: Review preserved vehicle-routing benchmark metadata, target outputs,
+    and metric definitions without executing the pipeline.
+- id: restore_first
+  label: Restore First
+  description: Restore missing MoSES training implementation, routing execution code,
+    and LoRA expert components from upstream repository before treating this benchmark
+    as runnable.
+dataset_download:
+  primary_method: mixed
+  sources:
+  - url: https://github.com/panyxy/moses_vrp
+    type: upstream_repository
+  - url: https://deepscientist.cc/AISB/087_moses
+    type: local_archive
+  notes:
+  - The upstream GitHub repository (panyxy/moses_vrp) contains the official MoSES
+    implementation and should serve as the primary source for restoration.
+  - The current local package does not preserve an executable benchmark route; treat
+    dataset setup as part of upstream restoration rather than a bundled run path.
+  - VRP instance generation and evaluation datasets require separate acquisition from
+    the upstream repository.
+credential_requirements:
+  mode: none
+  items: []
+  notes:
+  - No authentication required for public access to upstream repository or local archive.
+resources:
+  minimum:
+    cpu_cores: 16
+    ram_gb: 64
+    disk_gb: 150
+    gpu_count: 1
+    gpu_vram_gb: 24
+    notes: Minimum configuration suitable for single-expert inference and basic analysis.
+  recommended:
+    cpu_cores: 32
+    ram_gb: 128
+    disk_gb: 300
+    gpu_count: 2
+    gpu_vram_gb: 48
+    notes: Recommended for multi-expert training with LoRA adapters and batch evaluation.
+environment:
+  python: '3.10'
+  cuda: '12.1'
+  pytorch: 2.6.0
+  flash_attn: 2.7.0
+  key_packages:
+  - torch
+  - numpy
+  - scipy
+  - networkx
+  - pyyaml
+  notes:
+  - The current local package does not preserve executable setup files or runnable
+    source code; environment hints here should be treated as provisional until upstream
+    restoration.
+  - See the bundled README/requirements for the full dependency set once restoration
+    is complete.
+  - LoRA implementation typically requires flash_attn for efficient attention computation
+    during training.
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- no_executable_anchor
+risk_notes:
+- The current snapshot contains no executable code files that can anchor the staged
+  metrics; all primary outputs (optimality_gap_pct, tour_cost, inference_time) are
+  blocked from execution.
+- The local package preserves only metadata and paper artifacts; restoration from
+  the upstream GitHub repository is required before benchmark execution.
+- No benchmark execution was performed in this packaging pass.
+- The top-level README is absent in the current snapshot; consult upstream documentation
+  for operational details.
+recommended_when: Use this entry when you want paper-backed planning, upstream restoration
+  for the MoSES multi-stage routing pipeline, or analysis of the State-Decomposable
+  MDP framework and LoRA expert composition approach.
+not_recommended_when: Do not treat this local package as a ready-to-run routing benchmark;
+  the packaged snapshot does not preserve the executable training pipeline or evaluation
+  code. Do not attempt benchmark execution without first restoring the missing components
+  from the upstream repository.
+paper:
+  title: Multi-Task Vehicle Routing Solver via Mixture of Specialized Experts under
+    State-Decomposable MDP
+  authors:
+  - Yuxin Pan (HKUST)
+  - Zhiguang Cao (SMU)
+  - Chengyang Gu (HKUST-GZ)
+  - Liu Liu (Tencent AI Lab)
+  - Peilin Zhao (Tencent AI Lab, SJTU)
+  - Yize Chen (U Alberta)
+  - Fangzhen Lin (HKUST)
+  venue: NeurIPS
+  year: 2025
+  url: https://openreview.net/forum?id=ezSyZM6Lp7
+  code_url: https://github.com/panyxy/moses_vrp
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.087_moses.zip
+  archive_type: zip
+  local_dir_name: paper-87-MoSES
+  upstream_url: https://github.com/panyxy/moses_vrp
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.087_moses.zip
+  sha256: 26adaf9a54f2476f1ec7e293992e40065c0f6fb02abe3b702cd794bb07730c69
+  size_bytes: 59011
+display:
+  palette_seed: orange-navy-routing
+  art_style: optimization-playbook
+  accent_priority: high
+  tags:
+  - vehicle-routing
+  - mixture-of-experts
+  - lora
+  - multi-task-learning
+  - combinatorial-optimization
+  - state-decomposable-mdp
+image_path: ../image/087_aisb.t3.087_moses.jpg
+capability_tags:
+- research_code_optimization
+- combinatorial_optimization
+- vehicle_routing
+- mixture_of_experts
+- multi_task_learning
+- lora_specialization
+- latent_space_policy
+aisb_direction: T3
+track_fit:
+- paper_track
+- benchmark_track

package/AISB/catalog/aisb.t3.087_moses.zh.yaml

@@ -0,0 +1,132 @@
+id: aisb.t3.087_moses
+name: MoSES - 多任务车辆路径求解器
+version: 0.1.0
+one_line: 使用基于 LoRA 的专业专家和状态可分解 MDP 框架下的自适应门控的多任务车辆路径求解器。
+task_description: 该打包基准测试涵盖了用于解决多任务车辆路径问题的专业专家混合求解器 (MoSES)。该方法引入了状态可分解 MDP (SDMDP)，通过将状态空间表示为基础状态空间的笛卡尔积来重新表述 VRP，每个基础状态空间都与一个基础 VRP 变体相关联。基础策略是通过专门的 LoRA 专家实现的，每个专家都根据特定基础变体的冻结预训练骨干进行微调。潜在空间扩展（LS-SDMDP）结合了最佳基础策略和可学习的混合函数，以实现嵌入空间中的策略重用。混合函数通过自适应门控机制将最优基础策略生成的基础状态嵌入映射到统一状态嵌入。在温和的假设下，LS-SDMDP 可证明恢复最优统一策略。当前的本地快照保留基准元数据和论文工件，但不包含可执行的训练或评估代码；在执行基准测试之前需要恢复可运行管道。
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 1d+
+cost_band: high
+difficulty: hard
+data_access: public
+primary_outputs:
+- optimality_gap_pct
+- tour_cost
+- inference_time
+launch_profiles:
+- id: analysis_only
+  label: 仅分析
+  description: 在不执行管道的情况下查看保留的车辆路线基准元数据、目标输出和指标定义。
+- id: restore_first
+  label: 首先恢复
+  description: 在将该基准测试视为可运行之前，从上游存储库恢复丢失的 MoSES 训练实施、路由执行代码和 LoRA 专家组件。
+dataset_download:
+  primary_method: mixed
+  sources:
+  - url: https://github.com/panyxy/moses_vrp
+    type: upstream_repository
+  - url: https://deepscientist.cc/AISB/087_moses
+    type: local_archive
+  notes:
+  - 上游 GitHub 存储库 (panyxy/moses_vrp) 包含官方 MoSES 实现，应作为恢复的主要来源。
+  - 当前本地包不保留可执行的基准测试路线；将数据集设置视为上游恢复的一部分，而不是捆绑的运行路径。
+  - VRP实例生成和评估数据集需要从上游存储库单独获取。
+credential_requirements:
+  mode: none
+  items: []
+  notes:
+  - 公共访问上游存储库或本地存档不需要身份验证。
+resources:
+  minimum:
+    cpu_cores: 16
+    ram_gb: 64
+    disk_gb: 150
+    gpu_count: 1
+    gpu_vram_gb: 24
+    notes: 适合单专家推理和基础分析的最低配置。
+  recommended:
+    cpu_cores: 32
+    ram_gb: 128
+    disk_gb: 300
+    gpu_count: 2
+    gpu_vram_gb: 48
+    notes: 推荐用于 LoRA 适配器的多专家培训和批量评估。
+environment:
+  python: '3.10'
+  cuda: '12.1'
+  pytorch: 2.6.0
+  flash_attn: 2.7.0
+  key_packages:
+  - torch
+  - numpy
+  - scipy
+  - networkx
+  - pyyaml
+  notes:
+  - 当前本地包不保留可执行安装文件或可运行源代码；这里的环境提示应被视为临时的，直到上游恢复为止。
+  - 恢复完成后，请参阅捆绑的自述文件/要求以获取完整的依赖项集。
+  - LoRA 实现通常需要 flash_attn 来在训练期间进行有效的注意力计算。
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- no_executable_anchor
+risk_notes:
+- 当前快照不包含可以锚定暂存指标的可执行代码文件；所有主输出（optimality_gap_pct、tour_cost、inference_time）都被阻止执行。
+- 本地包仅保留元数据和纸质工件；在执行基准测试之前，需要从上游 GitHub 存储库进行恢复。
+- 在此打包过程中未执行任何基准测试。
+- 当前快照中缺少顶级自述文件；有关操作细节，请参阅上游文档。
+recommended_when: 当您需要纸质规划、MoSES 多级路由管道的上游恢复或状态可分解 MDP 框架和 LoRA 专家组合方法的分析时，请使用此条目。
+not_recommended_when: 不要将此本地包视为可立即运行的路由基准；打包的快照不保留可执行的训练管道或评估代码。在未先从上游存储库恢复丢失的组件之前，请勿尝试执行基准测试。
+paper:
+  title: Multi-Task Vehicle Routing Solver via Mixture of Specialized Experts under State-Decomposable MDP
+  authors:
+  - Yuxin Pan (HKUST)
+  - Zhiguang Cao (SMU)
+  - Chengyang Gu (HKUST-GZ)
+  - Liu Liu (Tencent AI Lab)
+  - Peilin Zhao (Tencent AI Lab, SJTU)
+  - Yize Chen (U Alberta)
+  - Fangzhen Lin (HKUST)
+  venue: NeurIPS
+  year: 2025
+  url: https://openreview.net/forum?id=ezSyZM6Lp7
+  code_url: https://github.com/panyxy/moses_vrp
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.087_moses.zip
+  archive_type: zip
+  local_dir_name: paper-87-MoSES
+  upstream_url: https://github.com/panyxy/moses_vrp
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.087_moses.zip
+  sha256: 26adaf9a54f2476f1ec7e293992e40065c0f6fb02abe3b702cd794bb07730c69
+  size_bytes: 59011
+display:
+  palette_seed: orange-navy-routing
+  art_style: optimization-playbook
+  accent_priority: high
+  tags:
+  - vehicle-routing
+  - mixture-of-experts
+  - lora
+  - multi-task-learning
+  - combinatorial-optimization
+  - state-decomposable-mdp
+image_path: ../image/087_aisb.t3.087_moses.jpg
+capability_tags:
+- research_code_optimization
+- combinatorial_optimization
+- vehicle_routing
+- mixture_of_experts
+- multi_task_learning
+- lora_specialization
+- latent_space_policy
+aisb_direction: T3
+track_fit:
+- paper_track
+- benchmark_track

package/AISB/catalog/aisb.t3.088_channelnorm.yaml

@@ -0,0 +1,140 @@
+id: aisb.t3.088_channelnorm
+name: Channel Normalization for Time Series Channel Identification
+version: 0.1.0
+one_line: Channel-specific affine normalization to enhance channel identifiability
+  in time-series forecasting models.
+task_description: 'This benchmark entry covers Channel Normalization (CN), a normalization
+  strategy that assigns distinct affine transformation parameters to each channel
+  in time-series models, enabling channel identifiability (CID). The current local
+  snapshot preserves paper metadata and latex source but does not contain executable
+  benchmark code or runnable evaluation scripts. The packaged benchmark is intended
+  for paper-backed planning and upstream restoration of the CN forecasting implementation.
+  Two variants are proposed: Adaptive CN (ACN) dynamically adjusts parameters based
+  on input time series, while Prototypical CN (PCN) uses learnable prototypes to handle
+  datasets with unknown or varying numbers of channels, facilitating use in time-series
+  foundation models.
+
+  '
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 2-6h
+cost_band: medium
+difficulty: medium
+data_access: public
+primary_outputs:
+- mse
+- channel_identifiability_metrics
+- forecasting_accuracy
+launch_profiles:
+- id: analysis_only
+  label: Analysis Only
+  description: Review preserved channel-normalization benchmark metadata, metric definitions,
+    and target MSE evaluation criteria.
+- id: restore_first
+  label: Restore First
+  description: Restore the missing ChannelNorm forecasting implementation from upstream
+    before attempting benchmark execution. Upstream restoration required for executable
+    code paths.
+dataset_download:
+  primary_method: restoration_required
+  sources: []
+  notes:
+  - The current local package does not preserve an executable benchmark route; dataset
+    and code setup are part of upstream restoration rather than a bundled run path.
+  - The paper references standard time-series benchmarks (ETTm1, Weather, PEMS03)
+    but these are not bundled in the current snapshot.
+credential_requirements:
+  mode: none
+  items: []
+  notes: []
+resources:
+  minimum:
+    cpu_cores: 8
+    ram_gb: 16
+    disk_gb: 50
+    gpu_count: 1
+    gpu_vram_gb: 8
+  recommended:
+    cpu_cores: 16
+    ram_gb: 32
+    disk_gb: 100
+    gpu_count: 1
+    gpu_vram_gb: 16
+environment:
+  python: '3.10'
+  cuda: '11.8'
+  pytorch: 2.1.0
+  flash_attn: null
+  key_packages: []
+  notes:
+  - The current local package does not preserve executable setup files or runnable
+    source code; environment specifications here are provisional based on typical
+    time-series deep learning setups.
+  - The paper applies CN to various backbones including Transformers (iTransformer),
+    MLPs (RMLP, TSMixer), and SSMs (Mamba variants).
+  - Treat environment hints as pending upstream restoration; see bundled README/requirements
+    for full dependency set when available.
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- no_executable_anchor
+risk_notes:
+- The current snapshot is blocked; no executable code files detected that can anchor
+  staged metrics.
+- No benchmark execution was performed in this packaging pass.
+- Primary metric (mse) is blocked-no-code in the current snapshot.
+- Top-level README is absent in the current snapshot.
+- Do not treat this local package as a ready-to-run normalization benchmark.
+recommended_when: 'Use this entry when planning paper-backed research on channel-identifiable
+  time-series models, evaluating upstream restoration candidates for the ChannelNorm
+  forecasting route, or analyzing normalization strategies for time-series forecasting
+  architectures.
+
+  '
+not_recommended_when: 'Do not use when seeking a ready-to-run normalization benchmark;
+  the staged snapshot preserves metadata only. Do not attempt execution without prior
+  upstream restoration of missing code.
+
+  '
+paper:
+  title: Channel Normalization for Time Series Channel Identification
+  venue: ICML 2025
+  year: 2025
+  url: https://arxiv.org/abs/2506.00432
+  code_url: https://github.com/seunghan96/CN
+  authors:
+  - Seunghan Lee
+  - Taeyoung Park
+  - Kibok Lee
+  institution: Yonsei University
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.088_channelnorm.zip
+  archive_type: zip
+  local_dir_name: paper-88-ChannelNorm
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.088_channelnorm.zip
+  sha256: bb1b5c0c7b3d2731d2698c1b4b9ec11f9601967281315c37e6548078eeb28913
+  size_bytes: 39531
+display:
+  palette_seed: cyan-slate-channel
+  art_style: normalization-grid
+  accent_priority: medium
+image_path: ../image/088_aisb.t3.088_channelnorm.jpg
+capability_tags:
+- research_code_optimization
+- time_series
+- normalization_layers
+- forecasting
+- channel_modeling
+track_fit:
+- paper_track
+- benchmark_track
+aisb_direction: T3
+commercial:
+  annual_fee: null

package/AISB/catalog/aisb.t3.088_channelnorm.zh.yaml

@@ -0,0 +1,109 @@
+id: aisb.t3.088_channelnorm
+name: 用于时间序列通道识别的通道归一化
+version: 0.1.0
+one_line: 特定于通道的仿射归一化可增强时间序列预测模型中的通道可识别性。
+task_description: 该基准条目涵盖了通道归一化 (CN)，这是一种归一化策略，可为时间序列模型中的每个通道分配不同的仿射变换参数，从而实现通道可识别性 (CID)。当前的本地快照保留纸张元数据和乳胶源，但不包含可执行的基准代码或可运行的评估脚本。打包基准旨在用于 CN 预测实施的纸质规划和上游恢复。提出了两种变体：自适应 CN (ACN) 根据输入时间序列动态调整参数，而原型 CN (PCN) 使用可学习原型来处理具有未知或变化通道数的数据集，便于在时间序列基础模型中使用。
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 2-6h
+cost_band: medium
+difficulty: medium
+data_access: public
+primary_outputs:
+- mse
+- channel_identifiability_metrics
+- forecasting_accuracy
+launch_profiles:
+- id: analysis_only
+  label: 仅分析
+  description: 查看保留的通道标准化基准元数据、指标定义和目标 MSE 评估标准。
+- id: restore_first
+  label: 首先恢复
+  description: 在尝试执行基准测试之前，从上游恢复丢失的 ChannelNorm 预测实现。可执行代码路径所需的上游恢复。
+dataset_download:
+  primary_method: restoration_required
+  sources: []
+  notes:
+  - 当前本地包不保留可执行的基准测试路线；数据集和代码设置是上游恢复的一部分，而不是捆绑的运行路径。
+  - 该论文引用了标准时间序列基准（ETTm1、Weather、PEMS03），但这些基准未捆绑在当前快照中。
+credential_requirements:
+  mode: none
+  items: []
+  notes: []
+resources:
+  minimum:
+    cpu_cores: 8
+    ram_gb: 16
+    disk_gb: 50
+    gpu_count: 1
+    gpu_vram_gb: 8
+  recommended:
+    cpu_cores: 16
+    ram_gb: 32
+    disk_gb: 100
+    gpu_count: 1
+    gpu_vram_gb: 16
+environment:
+  python: '3.10'
+  cuda: '11.8'
+  pytorch: 2.1.0
+  flash_attn: null
+  key_packages: []
+  notes:
+  - 当前本地包不保留可执行安装文件或可运行源代码；这里的环境规范是基于典型时间序列深度学习设置的临时环境规范。
+  - 该论文将 CN 应用于各种骨干网，包括 Transformer (iTransformer)、MLP（RMLP、TSMixer）和 SSM（Mamba 变体）。
+  - 将环境提示视为待处理的上游恢复；请参阅捆绑的自述文件/要求以获取完整的依赖项集（如果可用）。
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- no_executable_anchor
+risk_notes:
+- 当前快照被阻止；未检测到可以锚定分阶段指标的可执行代码文件。
+- 在此打包过程中未执行任何基准测试。
+- 主要指标 (mse) 在当前快照中是无代码阻塞的。
+- 当前快照中缺少顶级自述文件。
+- 不要将此本地包视为准备运行的规范化基准。
+recommended_when: 在规划对通道可识别时间序列模型的论文支持研究、评估 ChannelNorm 预测路线的上游恢复候选者或分析时间序列预测架构的标准化策略时，请使用此条目。
+not_recommended_when: 在寻求可立即运行的标准化基准时不要使用；暂存快照仅保留元数据。在没有事先上游恢复丢失代码的情况下，不要尝试执行。
+paper:
+  title: Channel Normalization for Time Series Channel Identification
+  venue: ICML 2025
+  year: 2025
+  url: https://arxiv.org/abs/2506.00432
+  code_url: https://github.com/seunghan96/CN
+  authors:
+  - Seunghan Lee
+  - Taeyoung Park
+  - Kibok Lee
+  institution: Yonsei University
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.088_channelnorm.zip
+  archive_type: zip
+  local_dir_name: paper-88-ChannelNorm
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.088_channelnorm.zip
+  sha256: bb1b5c0c7b3d2731d2698c1b4b9ec11f9601967281315c37e6548078eeb28913
+  size_bytes: 39531
+display:
+  palette_seed: cyan-slate-channel
+  art_style: normalization-grid
+  accent_priority: medium
+image_path: ../image/088_aisb.t3.088_channelnorm.jpg
+capability_tags:
+- research_code_optimization
+- time_series
+- normalization_layers
+- forecasting
+- channel_modeling
+track_fit:
+- paper_track
+- benchmark_track
+aisb_direction: T3
+commercial:
+  annual_fee: null