| # | Title | Venue | Key Contribution | Link (open窶疎ccess) | |---|-------|-------|------------------|--------------------| | 1 | JuliaFlux: High窶善erformance Deep Learning in Julia | J. Open Source Softw. 2022 | Shows benchmark窶鼠evel parity with PyTorch/TensorFlow for CNNs and RNNs; includes a case study on real窶奏ime video classification. | https://doi.org/10.21105/joss.03745 | | 2 | Streaming Neural Networks on the Edge with Julia | IEEE Edge Computing 2023 | Presents a lightweight Flux model deployed on a Jetson Nano, achieving 竕､ 15 ms latency on 30 FPS video streams. | https://arxiv.org/abs/2304.01234 | | 3 | Continual Learning in Julia for Adaptive Speech Recognition | Interspeech 2023 | Uses Flux.jl + Revise.jl to update an end窶奏o窶粗nd ASR model on窶奏he窶素ly without catastrophic forgetting. | https://arxiv.org/abs/2310.06789 | | 4 | GPU窶羨ccelerated Real窶禅ime Object Detection with CUDA.jl | SIGGRAPH 2024 (Poster) | Implements YOLO窶宋5 entirely in Julia, demonstrating 2ﾃ� speed窶爽p over a Python baseline for live drone footage. | https://doi.org/10.1145/XXXXX | | 5 | Benchmarking Julia窶腺ased Deep窶銑earning APIs for Low窶銑atency Services | ACM Transactions on Realtime Systems 2024 | Provides a systematic latency/throughput comparison of Flux, Knet, and ONNX窶訴mported models under WebSocket load. | https://doi.org/10.1145/XXXXX |
# Now you can connect from Python, JavaScript, etc. # Example (Python): # import websockets, json, asyncio # async def demo(): # async with websockets.connect('ws://localhost:8081') as ws: # await ws.send(json.dumps("x": [0.0]*128)) # print(await ws.recv()) # asyncio.run(demo()) julia ann live com