Hardware accelerators, especially those designed for tensor processing, have become ubiquitous in today's computing landscape. However, even with significant efforts in building compilers, programming these tensor accelerators remains challenging, leaving much of their potential underutilized. Recently, large language models (LLMs), trained on large amounts of code, have shown significant promise in code generation and optimization tasks, but generating low-resource languages, such as specialized tensor accelerator code still poses a significant challenge. We tackle this challenge with Autocomp, an approach that empowers accelerator programmers to leverage domain knowledge and hardware feedback to optimize code via an automated LLM-driven search. We accomplish this by: 1) formulating each optimization pass as a structured two-phase prompt, divided into planning and code generation phases, 2) inserting domain knowledge during planning via a concise and adaptable optimization menu, and 3) integrating correctness and performance metrics from hardware as feedback at each search iteration. Across three distinct hardware platforms, we demonstrate that Autocomp-optimized code runs 5.6x faster than the vendor-provided library (Gemmini), outperforms expert-level hand-tuned code by 1.9x (AWS Trainium), and achieves 3.8x higher performance than a machine learning-based cost model for GPUs (NVIDIA L40S). Additionally, we demonstrate that optimization schedules generated from Autocomp can be reused across similar tensor operations, improving speedups by up to 24% under a fixed sample budget.

To meet the increasing demand of deep learning (DL) models, AI chips are employing both off-chip memory (e.g., HBM) and high-bandwidth low-latency interconnect for direct inter-core data exchange. However, it is not easy to explore the efficiency of these inter-core connected AI (ICCA) chips, due to a fundamental tussle among compute (per-core execution), communication (inter-core data exchange), and I/O (off-chip data access). In this paper, we develop Elk, a DL compiler framework to maximize the efficiency of ICCA chips by jointly trading off all the three performance factors discussed above. Elk structures these performance factors into configurable parameters and forms a global trade-off space in the DL compiler. To systematically explore this space and maximize overall efficiency, Elk employs a new inductive operator scheduling policy and a cost-aware on-chip memory allocation algorithm. It generates globally optimized execution plans that best overlap off-chip data loading and on-chip execution. To examine the efficiency of Elk, we build a full-fledged emulator based on a real ICCA chip IPU-POD4, and an ICCA chip simulator for sensitivity analysis with different interconnect network topologies. Elk achieves 94% of the ideal roofline performance of ICCA chips on average, showing the benefits of supporting large DL models on ICCA chips. We also show Elk's capability of enabling architecture design space exploration for new ICCA chip development.

Intuitively, prostheses with user-controllable stiffness could mimic the intrinsic behavior of the human musculoskeletal system, promoting safe and natural interactions and task adaptability in real-world scenarios. However, prosthetic design often disregards compliance because of the additional complexity, weight, and needed control channels. This paper focuses on designing a Variable Stiffness Actuator (VSA) with weight, size, and performance compatible with prosthetic applications, addressing its implementation for the elbow joint. While a direct biomimetic approach suggests adopting an Agonist-Antagonist (AA) layout to replicate the biceps and triceps brachii with elastic actuation, this solution is not optimal to accommodate the varied morphologies of residual limbs. Instead, we employed the AA layout to craft an elbow prosthesis fully contained in the user's forearm, catering to individuals with distal transhumeral amputations. Additionally, we introduce a variant of this design where the two motors are split in the upper arm and forearm to distribute mass and volume more evenly along the bionic limb, enhancing comfort for patients with more proximal amputation levels. We characterize and validate our approach, demonstrating that both architectures meet the target requirements for an elbow prosthesis. The system attains the desired 120{\deg} range of motion, achieves the target stiffness range of [2, 60] Nm/rad, and can actively lift up to 3 kg. Our novel design reduces weight by up to 50% compared to existing VSAs for elbow prostheses while achieving performance comparable to the state of the art. Case studies suggest that passive and variable compliance could enable robust and safe interactions and task adaptability in the real world.

This paper presents a comprehensive numerical analysis of centrifugal clutch systems integrated with a two-speed automatic transmission, a key component in automotive torque transfer. Centrifugal clutches enable torque transmission based on rotational speed without external controls. The study systematically examines various clutch configurations effects on transmission dynamics, focusing on torque transfer, upshifting, and downshifting behaviors under different conditions. A Deep Neural Network (DNN) model predicts clutch engagement using parameters such as spring preload and shoe mass, offering an efficient alternative to complex simulations. The integration of deep learning and numerical modeling provides critical insights for optimizing clutch designs, enhancing transmission performance and efficiency.

Discriminatively trained, deterministic neural networks are the de facto choice for classification problems. However, even though they achieve state-of-the-art results on in-domain test sets, they tend to be overconfident on out-of-distribution (OOD) data. For instance, ReLU networks -- a popular class of neural network architectures -- have been shown to almost always yield high confidence predictions when the test data are far away from the training set, even when they are trained with OOD data. We overcome this problem by adding a term to the output of the neural network that corresponds to the logit of an extra class, that we design to dominate the logits of the original classes as we move away from the training data.This technique provably prevents arbitrarily high confidence on far-away test data while maintaining a simple discriminative point-estimate training. Evaluation on various benchmarks demonstrates strong performance against competitive baselines on both far-away and realistic OOD data.

Nonsurgical treatment of Dropped Head Syndrome (DHS) incurs the use of collar-type orthoses that immobilize the neck and cause discomfort and sores under the chin. Articulated orthoses have the potential to support the head posture while allowing partial mobility of the neck and reduced discomfort and sores. This work presents the design, modeling, development, and characterization of a novel multi-degree-of-freedom elastic mechanism designed for neck support. This new type of elastic mechanism allows the bending of the head in the sagittal and coronal planes, and head rotations in the transverse plane. From these articulate movements, the mechanism generates moments that restore the head and neck to the upright posture, thus compensating for the muscle weakness caused by DHS. The experimental results show adherence to the empirical characterization of the elastic mechanism under flexion to the model-based calculations. A neck support orthosis prototype based on the proposed mechanism is presented, which enables the three before-mentioned head motions of a healthy participant, according to the results of preliminary tests.

Left ventricular assist devices (LVADs) are mechanical pumps, which can be used to support heart failure (HF) patients as bridge to transplant and destination therapy. To automatically adjust the LVAD speed, a physiological control system needs to be designed to respond to variations of patient hemodynamics across a variety of clinical scenarios. These control systems require pressure feedback signals from the cardiovascular system. However, there are no suitable long-term implantable sensors available. In this study, a novel real-time deep convolutional neural network (CNN) for estimation of preload based on the LVAD flow was proposed. A new sensorless adaptive physiological control system for an LVAD pump was developed using the full dynamic form of model free adaptive control (FFDL-MFAC) and the proposed preload estimator to maintain the patient conditions in safe physiological ranges. The CNN model for preload estimation was trained and evaluated through 10-fold cross validation on 100 different patient conditions and the proposed sensorless control system was assessed on a new testing set of 30 different patient conditions across six different patient scenarios. The proposed preload estimator was extremely accurate with a correlation coefficient of 0.97, root mean squared error of 0.84 mmHg, reproducibility coefficient of 1.56 mmHg, coefficient of variation of 14.44 %, and bias of 0.29 mmHg for the testing dataset. The results also indicate that the proposed sensorless physiological controller works similarly to the preload-based physiological control system for LVAD using measured preload to prevent ventricular suction and pulmonary congestion. This study shows that the LVADs can respond appropriately to changing patient states and physiological demands without the need for additional pressure or flow measurements.