How spaicom/semantic-alignment-via-dual-sim works
This repository implements research experiments for over-the-air semantic alignment: prepare vision embeddings, train or select an alignment strategy, transmit through simulated SIM stacks and wireless channels, then score classification or reconstruction quality. It also supports learned unrolled optimization for channel-varying joint Dual-SIM configurations and plotting experiment results. The graph emphasizes runtime experiment paths and shared simulation layers. README statements and indexed references support the shown links; several script and model internals were not sampled, so uncertain implementation-level connections are omitted.
Main entry points: classification.py Classification Runs · tasks.py Task Scoring · reconstruction.py Reconstruction Runs · joint.py Joint Dual-SIM
Components 24
Data and Tasks
- Data Preparation
src/semantic_alignment/data/prepare.py - Semantic Embeddings
src/semantic_alignment/data/semasia.py - Reconstruction Data
src/semantic_alignment/data/reconstruction.py - Autoencoders
src/semantic_alignment/models/autoencoder.py - Task Scoring
src/semantic_alignment/evaluation/tasks.py - Link Inference
src/semantic_alignment/evaluation/inference.py - Linear Probe
src/semantic_alignment/evaluation/linear_probe.py
SIM Physical Layer
- Channel Utilities
src/semantic_alignment/utils/channel.py - SIM Architectures
src/semantic_alignment/sim/original - Joint Dual-SIM
src/semantic_alignment/sim/original/joint.py - Mono-SIM Optimizer
src/semantic_alignment/sim/original/optimizers/monosim.py - Dual-SIM Optimizer
src/semantic_alignment/sim/original/optimizers/dualsim.py - Cascade Simulation
src/semantic_alignment/sim/_cascade.py - SCA Optimization
src/semantic_alignment/sim/core.py - Transmit Link
src/semantic_alignment/sim/link.py
Unrolled Optimization
- Unrolled Runs
scripts/unrolled_classification.py - Unrolled Pipeline
scripts/unrolled_pipeline.py - Learned Steps
src/semantic_alignment/sim/unrolled/_step.py - Unrolled Runner
src/semantic_alignment/sim/unrolled/runner.py
Experiments and Results
- Classification Runs
scripts/classification.py - Reconstruction Runs
scripts/reconstruction.py - Classification Plots
scripts/plot_results_classification.py - Reconstruction Plots
scripts/plot_results_reconstruction.py
Other components
- Researcher
Connections 27
- Researcher → Classification Runs: launches
README.md - Researcher → Reconstruction Runs: launches
README.md - Researcher → Unrolled Runs: launches
README.md - Classification Runs → Data Preparation: loads data
scripts/classification.py - Data Preparation → Semantic Embeddings: uses embeddings
src/semantic_alignment/data/prepare.py - Classification Runs → Linear Probe: evaluates
scripts/classification.py - Classification Runs → Task Scoring: configures task
scripts/classification.py - Classification Runs → Channel Utilities: uses channel tools
scripts/classification.py - Reconstruction Runs → Reconstruction Data: uses data
README.md - Reconstruction Data → Autoencoders: uses models
src/semantic_alignment/data/reconstruction.py - Reconstruction Runs → Task Scoring: scores runs
README.md - Classification Runs → SIM Architectures: sweeps architectures
README.md - SIM Architectures → Joint Dual-SIM: includes joint
README.md - Joint Dual-SIM → Dual-SIM Optimizer: optimizes
src/semantic_alignment/sim/original/joint.py - Joint Dual-SIM → Transmit Link: builds transmit
src/semantic_alignment/sim/original/joint.py - Mono-SIM Optimizer → SCA Optimization: uses optimization
src/semantic_alignment/sim/original/optimizers/monosim.py - Dual-SIM Optimizer → SCA Optimization: uses optimization
src/semantic_alignment/sim/original/optimizers/dualsim.py - SCA Optimization → Cascade Simulation: builds cascades
src/semantic_alignment/sim/core.py - Transmit Link → Channel Utilities: adds noise
src/semantic_alignment/sim/link.py - Task Scoring → Link Inference: evaluates links
src/semantic_alignment/evaluation/tasks.py - Task Scoring → Transmit Link: scores transmit
src/semantic_alignment/evaluation/tasks.py - Unrolled Runs → Unrolled Pipeline: runs pipeline
scripts/unrolled_classification.py - Unrolled Pipeline → Unrolled Runner: trains and evaluates
scripts/unrolled_pipeline.py - Unrolled Runner → Learned Steps: applies steps
- Unrolled Pipeline → Task Scoring: scores results
scripts/unrolled_pipeline.py - Classification Plots → Classification Runs: plots results
README.md - Reconstruction Plots → Reconstruction Runs: plots results
README.md