Upload technical_report_semantic.md with huggingface_hub

technical_report_semantic.md

@@ -378,16 +378,14 @@ main encoder/decoder remain purely convolutional.
 ### 4.3 Noisy Alignment
 
 Unlike standard representation alignment which operates on clean latents,
-we align **noisy** latent versions. The noise level τ is sampled from a
-Beta(2,2) distribution (concentrated around τ=0.5) using flow matching
-linear interpolation:
+we align **noisy** latent versions. The noise level \\(\tau\\) is sampled from a
+\\(\text{Beta}(2,2)\\) distribution (concentrated around \\(\tau = 0.5\\)) using
+flow matching linear interpolation:
 
-```
-z_noisy = (1 - τ) · z + τ · ε,    ε ~ N(0, I),    τ ~ Beta(2, 2)
-```
+$$z_\text{noisy} = (1 - \tau) \, z + \tau \, \varepsilon, \qquad \varepsilon \sim \mathcal{N}(0, I), \quad \tau \sim \text{Beta}(2, 2)$$
 
-The projection head receives both the noisy latents and the noise level τ
-(via its AdaLN conditioning). This trains the head to extract semantic
+The projection head receives both the noisy latents and the noise level
+\\(\tau\\) (via its AdaLN conditioning). This trains the head to extract semantic
 information even from partially corrupted latents, improving robustness
 for downstream diffusion models which operate on noised latent inputs.
 
@@ -580,6 +578,20 @@ z_sampled = posterior.sample()
 
 ---
 
+## Citation
+
+```bibtex
+@misc{semdisdiffae,
+  title   = {SemDisDiffAE: A Semantically Disentangled Diffusion Autoencoder with FCDM Blocks},
+  author  = {data-archetype},
+  year    = {2026},
+  month   = apr,
+  url     = {https://huggingface.co/data-archetype/semdisdiffae},
+}
+```
+
+---
+
 ## 9. Results
 
 ## 7. Results