STLT
A two-stage system for energy-aware distributed inference between low-Earth-orbit satellites and ground stations. Offline analysis narrows the search space of DNN split points, and lightweight online decisions dynamically select the optimal split point, optimizing the trade-off between computation and communication energy.
Timeline (last 90 days)
STLT is a two-stage system for energy-aware distributed inference between LEO satellites and ground stations, whose core idea is to decouple "search" from "decision".
STLT's first stage is performed by an offline analyzer, which characterizes DNN architecture features and identifies a set of favorable split points, thereby reducing the online search space.
STLT's second stage is an onboard algorithm that performs lightweight energy-aware split selection, dynamically choosing the optimal split from the precomputed set of split points according to the current environmental state.
STLT was evaluated on a real satellite telemetry dataset and reduces the split candidate search space by an average of 30.7%.
STLT reduces decision latency by an average of 25.1%.
STLT reduces energy consumption by an average of 27.5%.
Compared with state-of-the-art satellite-ground collaborative inference baselines, STLT delivers up to 20.1% more inference tasks while consuming 8.8% less overall energy.
By combining offline precomputation with lightweight online decision-making, STLT transforms the complex energy optimization problem into fast selection over a limited candidate set.
STLT's measured performance shows that proper split point management can significantly improve overall system efficiency without introducing heavy prediction overhead.
All Facts (9)
STLT is a two-stage system for energy-aware distributed inference between LEO satellites and ground stations, whose core idea is to decouple "search" from "decision".
50%UnverifiedSTLT's first stage is performed by an offline analyzer, which characterizes DNN architecture features and identifies a set of favorable split points, thereby reducing the online search space.
50%UnverifiedSTLT's second stage is an onboard algorithm that performs lightweight energy-aware split selection, dynamically choosing the optimal split from the precomputed set of split points according to the current environmental state.
50%UnverifiedSTLT was evaluated on a real satellite telemetry dataset and reduces the split candidate search space by an average of 30.7%.
50%UnverifiedSTLT reduces decision latency by an average of 25.1%.
50%UnverifiedSTLT reduces energy consumption by an average of 27.5%.
50%UnverifiedCompared with state-of-the-art satellite-ground collaborative inference baselines, STLT delivers up to 20.1% more inference tasks while consuming 8.8% less overall energy.
50%UnverifiedBy combining offline precomputation with lightweight online decision-making, STLT transforms the complex energy optimization problem into fast selection over a limited candidate set.
50%UnverifiedSTLT's measured performance shows that proper split point management can significantly improve overall system efficiency without introducing heavy prediction overhead.
50%