A modular experimental framework for face transformation research, supporting simultaneous manipulation of multiple facial attributes (age, gender, etc.) using StyleGAN latent space editing.

The machinery behind machine learning is growing more complex, but the bottleneck often shows up in small, unglamorous work: ...

An agent built on Claude Code for formal theorem proving tasks. We used this system to prove all 12 problems from Putnam 2025, and completed a paper-level formalization of Effective Brascamp-Lieb ...

点击上方“Deephub Imba”,关注公众号,好文章不错过 !这篇文章从头实现 LLM-JEPA: Large Language Models Meet Joint Embedding Predictive Architectures。需要说明的是，这里写的是一个简洁的最小化训练脚本，目标是了解 JEPA 的本质：对同一文本创建两个视图，预测被遮蔽片段的嵌入，用表示对齐损失来训练。本文的目标是 ...

简单做一个个人作品集落地页：单页全屏 Hero 区域铺满荷兰弟的照片。鼠标移动时出现一个圆形聚光灯，像手电筒一样照亮第二张隐藏图片——形成 reveal mask 效果。快速甩鼠标拖出柔和残影，整个页面极简白底，交互和人物图像是唯一的视觉重心。