When drones learn to listen
📬 Distill AI delivers the most important AI papers worth your time, tailored to your field, every morning. 💬 Chat with any of them, instantly!
A four-drone swarm pulls secrets from embedded chips by sniffing electromagnetic emissions at 1.5m, Taylor series finds the surprising moments in long-form video, and a query-layer trick recovers fine detail from frozen vision models without retraining them. Plus three conference deadlines closing in the next three weeks.
⭐ TODAY’S HIGHLIGHT
TriSweep: A Four-Drone Swarm Framework for Electromagnetic Side-Channel Analysis
Three collector drones fan out around an embedded target while a coordinator drone orchestrates them, gathering electromagnetic emissions from 0.25 to 1.5 meters away to mount side-channel attacks. The paper sits at the intersection of swarm robotics, hardware security, and signal processing — a rare combination, and a real one to think about for anyone designing devices that assume physical access is the threat boundary. The whole pipeline runs at altitude.
📄 MORE PAPERS WORTH READING
• Swift Sampling: Selecting Temporal Surprises via Taylor Series — Identifies the highest-information frames in long-form video by detecting when visual features deviate from a predicted Taylor-series trajectory. Training-free, model-agnostic, drops in front of any downstream encoder.
• DecQ: Detail-Condensing Queries for Representation Autoencoders — A neat workaround for representation autoencoders that need fine-grained spatial detail but can’t afford to unfreeze the foundation model underneath. Improves both reconstruction and generation quality.
💻 ON GITHUB
• Octoday-Hub/Embodied-AI — 466 ⭐. Comprehensive knowledge index and industry mapping for embodied AI — companies, key papers, technical resources. A useful map of a field that’s moving fast.
• mimic-video/mimic-video — Python. Video-Action Models for robot control. Explores how video understanding generalizes robotic decision-making across tasks where vision-language policies fall short.
• ahammadmejbah/Awesome-Datasets-Hub — Curated LLM dataset directory spanning medical, NLP, multimodal, instruction-tuning, reasoning, code generation, and evaluation benchmarks. Worth bookmarking.
🤗 ON HUGGING FACE
• openai/whisper-large-v3-turbo — speech-recognition. Faster than Whisper v3 with comparable transcription accuracy across many languages. The new default for production ASR.
• bigcode/starcoder — code-generation. Trained on permissively licensed source across many languages — the open-licensed alternative when license-clean training data matters.
• meta-llama/Llama-2-7b-chat-hf — text-generation. 7B-parameter chat model — small enough to run anywhere, robust enough for serious dialogue work. Multi-turn context handled cleanly.
⏰ DEADLINES CLOSING SOON
• PRICAI 2026 — Pacific Rim International Conference on Artificial Intelligence — Guangzhou, China — submission deadline June 13, 2026 (20 days)
• ECCV 2026 — European Conference on Computer Vision — Malmö, Sweden — AI Art submission June 14, 2026 (21 days)
• WACV 2027 — IEEE/CVF Winter Conference on Applications of Computer Vision — Buena Vista, FL — Round 1 paper registration June 18, 2026 (25 days)