EDCompress optimizes energy efficiency of CNN models on edge devices.
problem Low energy consumption for edge devices with diverse dataflow types.
method Energy-aware model compression using reinforcement learning.
result Improves energy efficiency by 20X, 17X, 37X in various networks.
The paper proposes a dataflow-based approach to reduce energy consumption in deep neural networks without sacrificing performance.
problem Reducing energy consumption in deep neural networks without performance drop.
method Dataflow-based joint quantization of weights and activations.
result Joint quantization improves performance and reduces energy consumption.
Efficient method improves neural network device placement.
problem Challenging task of placing operations on suitable devices in neural networks.
method End-to-end scalable sequential attention mechanism over graph neural network.
result 16% improvement over human experts and 9.2% over prior art.
Adaptive financial dataflow system improves model robustness in dynamic markets.
problem Static historical data leads to poor performance in dynamic financial markets.
method Drift-aware dataflow system with adaptive control and optimization.
result Enhanced model robustness and improved risk-adjusted returns.
System uses neural networks to prove program equivalence via rewrite rules.
problem Proving equivalence between two dataflow graphs.
method Developed a graph-to-sequence neural network trained on example generation to find semantics-preserving rewrite rules.
result System correctly outputs a rewrite sequence for 96% of program pairs, proving equivalence.
This paper improves neural network training performance by optimizing concurrency and operation scheduling.
problem Managing and scheduling fine-grained operations in neural network training for high performance.
method Extending TensorFlow runtime to enable automatic concurrency control and scheduling, using performance modeling.
result Achieved 33% average performance improvement on neural network models, up to 49%.
Adds recursion to deep learning frameworks for better handling of recursive data structures.
problem Lack of support for recursion in existing deep learning frameworks.
method Complements existing frameworks with recursive execution of dataflow graphs and APIs for recursive definitions.
result Recursive implementation reduces training and inference time by more effectively using resources.
A new deep learning method using Boolean logic reduces training and inference energy.
problem High computational and energy costs in deep learning training and inference.
method Introduces Boolean weights and inputs for efficient training using Boolean logic.
result Achieves full-precision accuracy in ImageNet classification and surpasses state-of-the-art results in semantic segmentation.
Myia compiler optimizes ML models with efficient AD for array programming.
problem Efficient automatic differentiation for array programming in ML.
method Introduces a new graph-based IR that supports function calls, higher-order functions, and recursion.
result Myia compiler enables efficient AD using source transformation without a tape, supporting higher-order derivatives.
This work optimizes deep learning training by combining data and model parallelism.
problem Training large models with multiple GPUs suffers from high communication overhead and statistical efficiency loss.
method Hybrid parallelization combining data and model parallelism.
result Hybrid training provides significant speedup compared to data parallelism alone.
In recent years, there have been numerous developments towards solving multimodal tasks, aiming to learn a stronger representation than through a single modality. Certain aspects of the data can be particularly useful in this case - for example, correlations in the space or time domain across modalities - but should be…
Tiled Squeeze-and-Excite improves channel attention with local spatial context.
problem Improving channel attention mechanisms in neural networks.
method Proposes tiled squeeze-and-excite (TSE) framework for channel attention.
result Local context of 7 rows or columns is sufficient for matching global context performance.
Paper proposes NASAIC framework for co-designing neural architectures and heterogeneous ASICs.
problem Designing efficient neural architectures and ASICs for multiple tasks.
method Build ASIC templates and propose NASAIC framework for simultaneous design of architectures and ASICs.
result NASAIC ensures design specifications and maximizes accuracy with minimal performance loss.