Meta has traditionally relied on using CPU-based servers for running inference workloads, specifically Deep Learning Recommendation Models (DLRM), but the increasing compute and memory requirements of these models have pushed the company towards using specialized solutions such as GPUs or other hardware accelerators. This paper describes the company's effort in constructing its first silicon specifically designed for recommendation systems; it describes the accelerator architecture and platform design, the software stack for enabling and optimizing PyTorch-based models and provides an initial performance evaluation. With our emerging software stack, we have made significant progress towards reaching the same or higher efficiency as the GPU: We averaged 0.9x perf/W across various DLRMs, and benchmarks show operators such as GEMMs reaching 2x perf/W. Finally, the paper describes the lessons we learned during this journey which can improve the performance and programmability of future generations of architecture.

Predicting the motion of observed entities benefits humans almost seamlessly. The same benefits can be proliferated to mobile autonomous systems if we have a reliable, real-time solution to predict the motion of any object of interest, be it the host’s own motion or that of an observed foreign object. In this work, a novel Multi-Variable State Prediction (MVSP) methodology is devised for real-time trajectory prediction. MVSP incorporates cascaded stages of HMM with Viterbi algorithm and probabilistic quantization for accurately predicting the motion characteristics of the moving object. The overall scheme is employed to predict the motion of moving objects in a 3D space. The proposed approach is verified on both synthetically generated data sequences and data-sets captured from real-life experiments. For a practical scenario, the experiments resulted in an RMS error of 0.6m for a predicted distance of ~18m demonstrating the effectiveness and accuracy of the proposed methodology.

Optimal path planning and smooth trajectory planning are critical for effective navigation of mobile robots working towards accomplishing complex missions. For autonomous, real time and extended operations of mobile robots, the navigation capability needs to be executed at the edge. Thus, efficient compute, minimum memory utilization and smooth trajectory are the key parameters that drive the successful operation of autonomous mobile robots. Traditionally, navigation solutions focus on developing robust path planning algorithms which are complex and compute/memory intensive. Bidirectional-RRT(Bi-RRT) based path planning algorithms have gained increased attention due to their effectiveness and computational efficiency in generating feasible paths. However, these algorithms neither optimize memory nor guarantee smooth trajectories. To this end, we propose a kinematically constrained Bi-RRT (KB-RRT) algorithm, which restricts the number of nodes generated without compromising on the accuracy and incorporates kinodynamic constraints for generating smooth trajectories, together resulting in efficient navigation of autonomous mobile robots. The proposed algorithm is tested in a highly cluttered environment on an Ackermannsteering vehicle model with severe kinematic constraints. The experimental results demonstrate that KB-RRT achieves three times (3 X) better performance in terms of convergence rate and memory utilization compared to a standard Bi-RRT algorithm.

A Coupled Hidden Markov Model (CHMM) is proposed in this paper to perform multimodal speech recognition using myoeletric signals (MES) from the muscles of vocal articulation. MES signals are immune to noise, and words that are acoustically similar manifest distinctly in MES. Hence, they would effectively complement the acoustic data in a multimodal speech recognition system. Research in Audio-Visual Speech Recognition has shown that CHMMs model the asynchrony between different data streams effectively. Hence, we propose CHMM for multimodal speech recognition using audio and MES as the two data streams. Our experiments indicate that the multimodal CHMM system significantly outperforms the audio only system at different SNRs. We have also provided a comparison between different features for MES and have found that wavelet features provide the best results.