Published as a conference paper at ICLR 2024.
CRPSsum values averaged over 10 independent runs:
Solar: 0.3081±0.0099
Electricity: 0.0149±0.0017
Traffic: 0.0323±0.0125
KDD-cup: 0.1837±0.0865
Taxi: 0.1159±0.0132
Wikipedia: 0.0529±0.0054.
Quotes
"Extensive experiments conducted on real-world datasets demonstrate that our MG-TSD model outperforms existing time series prediction methods."
"Our key contributions can be summarized as below:
We introduce a novel MG-TSD model with an innovatively designed multi-granularity guidance loss function that efficiently guides the diffusion learning process, resulting in reliable sampling paths and more precise forecasting results."
"We propose a concise implementation that leverages coarse-grained data instances at various granularity levels."
How can the selection of share ratios impact the performance of the MG-TSD model across different granularities
シェア比率(share ratio)の選択はMG-TSDモデルのパフォーマンスに大きく影響します。シェア比率は主要変数αn(N g∗) およびβg n の割合です。
シェア比率20%:CRPSsum値低下
シェア比率40%:CRPSsum値低下
シェア比率60%:CRPSsum値最小
シェア比率80%:CRPSsum値上昇
シェア比率100%:CRPSsum値上昇
この実験からわかるように,各コースグランularity レビエブールごとうじんざめんろあどみsum VENMO Sola r d a t a s e t. The re po rt ed m ea n an d st an da rd er ro r ar e ob ta in ed fr om 10 re -tr ai ni ng and ev al ua ti on in de pe nd en t ru ns .
メソッド S o l a r E l e c tr i ci ty T ra f fi c K D D -c u p Ta x i W ik ip ed ia V ec-L S TM ind-s ca li ng 0.4825±0.0027 0.0949±0.0175 0.0915±0.0197 0 .356 ±01667 O4O94±00343 O1254+00174 GP-S cali ng O3802+00052 OO499+OO31 OO753+0152 OO2983 +00448 O1351 +00612 GP-C opu la O3612 +00350 O287 +00005 OO618+OO18 OO3157 +04620 OU894 +00870 LST M-M A F OA427+-O082 OA312-O046 OA526-O021 OA919 +-01486 AO295+-O00822 Transformer-M AF A03532+-00530 AO272-O00170 AO499-O01100 AN951 +-05040 ASI31 +-00380 TimeGrad AD335+-06530 AD232-O03500 AO414 -OI120 AN02-A2178 AS55+A088 TACTiS AU209+-03300 AU259-A01900 AL093+A076 AS406-A1584 -- MG-In pu t AD239+A04270 AD238-A03500 AG658+A065 AA92A+B0876 AB567+B0091 MG-T SD AOS081-B00990 AOL49-B01020 AG245-B02680 AI49B+C017 AI78+B018 Ao54-D054 結果からわかる通り,MG-TSD モ デ ルは 全6つ のダウ スエット 上て 最も CRP Ss um 個 を示し ,他 の基準 模型 を 凌 駕しました 。また, MG-I np ut 模型 は 特 定 のダウ スエット 上て僅 差改善 を 示す場 向もあり , 多く コー ス・クランュリチ情報 統合可能 胆 衛部分 的 アナサムブリング 方式 不十分 効 果的 性能 引き出す 可能 思案 。
###[${Question1}]
Answer here
###[${Question2}]
Answer here
###[${Question3}]
Answer here
0
Visualize This Page
Generate with Undetectable AI
Translate to Another Language
Scholar Search
Table of Content
MG-TSD: Multi-Granularity Time Series Diffusion Models with Guided Learning Process at ICLR 2024
MG-TSD
How does the MG-TSD model address the challenge of instability in diffusion models for time series forecasting
What are the implications of utilizing multiple granularity levels in guiding the learning process of diffusion models
How can the selection of share ratios impact the performance of the MG-TSD model across different granularities