On-device training is essential for neural networks (NNs) to continuously adapt to new online data, but can be time-consuming due to the device's limited computing power. To speed up on-device training, existing schemes select trainable NN portion offline or conduct unrecoverable selection at runtime, but the evolution of trainable NN portion is constrained and cannot adapt to the current need for training. Instead, runtime adaptation of on-device training should be fully elastic, i.e., every NN substructure can be freely removed from or added to the trainable NN portion at any time in training. In this paper, we present ElasticTrainer, a new technique that enforces such elasticity to achieve the required training speedup with the minimum NN accuracy loss. Experiment results show that ElasticTrainer achieves up to 3.5× more training speedup in wall-clock time and reduces energy consumption by 2×-3× more compared to the existing schemes, without noticeable accuracy loss.
Graphics Processing Unit (GPU) on smartphones is an effective target for hardware attacks. In this paper, we present a new side channel attack on mobile GPUs of Android smartphones, allowing an unprivileged attacker to eavesdrop the user's credentials, such as login usernames and passwords, from their inputs through on-screen keyboard. Our attack targets on Qualcomm Adreno GPUs and investigate the amount of GPU overdraw when rendering the popups of user's key presses of inputs. Such GPU overdraw caused by each key press corresponds to unique variations of selected GPU performance counters, from which these key presses can be accurately inferred. Experiment results from practical use on multiple models of Android smartphones show that our attack can correctly infer more than 80% of user's credential inputs, but incur negligible amounts of computing overhead and network traffic on the victim device. To counter this attack, this paper suggests mitigations of access control on GPU performance counters, or applying obfuscations on the values of GPU performance counters.
Smartphone assistants built on large language models (LLMs) today either drive an application's screen display or wait to be asked through chatting; neither is integrated into the event stream produced by the mobile OS and user-installed applications. We demonstrate MobileClaw, an Android system component that attaches an external LLM to the OS notification channel. A background service captures every posted notification and asks the model for recommended further actions; an interactive front end carries out the user's requests through built-in skills, operated through OS API calls rather than screen automation. The model is treated as an untrusted component: nothing it decides takes effect unless the user confirms it. Notification text and everything the app reads from the device pass a sanitizer before they leave the phone, and privileged operations run in a separate process behind an allowlist it re-checks itself. Demo attendees may fire synthetic or real notifications, manipulate local resources on the device, approve or reject suggested actions, and watch the model's reasoning trace in real time.