Dense 与 MoE 神经网络信息图

{
"type": "infographic comparison diagram",
"header": {
"title": "{argument name=\"main title\" default=\"Dense と MoE の違い\"}"
},
"layout": {
"structure": "two main columns separated by a central VS badge, with a split footer at the bottom",
"sections": [
{
"position": "left column",
"theme_color": "blue",
"header": "{argument name=\"left model name\" default=\"Dense モデル\"}",
"subtitle": "全てのニューロンが活性化",
"diagram": {
"type": "fully connected neural network",
"elements": [
"1 orange input node labeled 入力",
"4 hidden layers with 4, 5, 4, and 2 nodes respectively",
"nodes colored white, blue, and yellow",
"dense intersecting connection lines between all adjacent nodes"
]
},
"bullet_points": {
"count": 2,
"items": [
"{argument name=\"left bullet point\" default=\"全てのパラメータが使用\"}",
"計算コストが高い"
]
}
},
{
"position": "right column",
"theme_color": "orange",
"header": "{argument name=\"right model name\" default=\"MoE モデル\"}",
"subtitle": "一部の専門家が選択的に活性化",
"diagram": {
"type": "mixture of experts network",
"elements": [
"1 orange input node labeled 入力",
"3 rectangular blocks labeled Expert 1, Expert 2, Expert 3",
"1 yellow output node labeled 出力",
"branching arrows connecting input to experts, and experts to output"
]
},
"bullet_points": {
"count": 2,
"items": [
"{argument name=\"right bullet point\" default=\"一部のエキスパートのみ使用\"}",
"効率的でスケーラブル"
]
}
},
{
"position": "center",
"element": "red circular badge with text VS",
"connections": "blue arrow pointing left, orange arrow pointing right"
},
{
"position": "footer left",
"background": "light blue",
"text": "Dense: すべての層が常時稼働し全パラメータを使用",
"icon": "1 CPU chip graphic",
"label": "高い消費電力"
},
{
"position": "footer right",
"background": "light orange",
"text": "MoE: 必要なエキスパートのみを動員",
"icons": "2 circular graphics (orange arrows, blue lightning bolt)",
"label": "低コスト・高効率"
}
]
}
}