Low Rank Adaptation (LoRA) has gained massive attention in the recent generative AI research. One of the main advantages of LoRA is its ability to be fused with pretrained models, adding no overhead during inference. However, from a mobile deployment standpoint, we can either avoid inference overhead in the fused mode but lose the ability to switch adapters rapidly, or suffer significant (up to 30% higher) inference latency while enabling rapid switching in the unfused mode. LoRA also exhibits concept-loss when multiple adapters are used concurrently. In this paper, we propose Sparse High Rank Adapters (SHiRA), a new paradigm which incurs no inference overhead, enables rapid switching, and significantly reduces concept-loss. Specifically, SHiRA can be trained by directly tuning only 1-2% of the base model weights while leaving others unchanged. This results in a highly sparse adapter which can be switched directly in the fused mode. We further provide theoretical and empirical insights on how high sparsity in SHiRA can aid multi-adapter fusion by reducing concept loss. Our extensive experiments on LVMs and LLMs demonstrate that finetuning only a small fraction of the parameters in the base model significantly outperforms LoRA while enabling both rapid switching and multi-adapter fusion. Finally, we provide a latency- and memory-efficient SHiRA implementation based on Parameter-Efficient Finetuning (PEFT) Library which trains at nearly the same speed as LoRA while consuming up to 16% lower peak GPU memory, thus making SHiRA easy to adopt for practical use cases. To demonstrate rapid switching benefits during inference, we show that loading SHiRA on a base model can be 5x-16x faster than LoRA fusion on a CPU.

The deployment and training of neural networks on edge computing devices pose many challenges. The low memory nature of edge devices is often one of the biggest limiting factors encountered in the deployment of large neural network models. Tensor rematerialization or recompute is a way to address high memory requirements for neural network training and inference. In this paper we consider the problem of execution time minimization of compute graphs subject to a memory budget. In particular, we develop a new constraint programming formulation called Moccasin with only $O(n)$ integer variables, where $n$ is the number of nodes in the compute graph. This is a significant improvement over the works in the recent literature that propose formulations with $O(n^2)$ Boolean variables. We present numerical studies that show that our approach is up to an order of magnitude faster than recent work especially for large-scale graphs.

Recent works on machine learning for combinatorial optimization have shown that learning based approaches can outperform heuristic methods in terms of speed and performance. In this paper, we consider the problem of finding an optimal topological order on a directed acyclic graph (DAG) with focus on the memory minimization problem which arises in compilers. We propose an end-to-end machine learning based approach for topological ordering using an encoder-decoder framework. Our encoder is a novel attention based graph neural network architecture called \emph{Topoformer} which uses different topological transforms of a DAG for message passing. The node embeddings produced by the encoder are converted into node priorities which are used by the decoder to generate a probability distribution over topological orders. We train our model on a dataset of synthetically generated graphs called layered graphs. We show that our model outperforms, or is on-par, with several topological ordering baselines while being significantly faster on synthetic graphs with up to 2k nodes. We also train and test our model on a set of real-world computation graphs, showing performance improvements.

Smart phones with increased computation and sensing capabilities have enabled the growth of a new generation of applications which are organic and designed to react depending on the user contexts. These contexts typically define the personal, social, work and urban spaces of an individual and are derived from the underlying sensor measurements. The shared context streams therefore embed in them information, which when stitched together can reveal behavioral patterns and possible sensitive inferences, raising serious privacy concerns. In this paper, we propose a model based technique to capture the relationship between these contexts, and better understand the privacy implications of sharing them. We further demonstrate that by using a generative model of the context streams we can simultaneously meet the utility objectives of the context-aware applications while maintaining individual privacy. We present our current implementation which uses offline model learning with online inferencing performed on the smart phone. Preliminary results are presented to provide proof-of-concept of our proposed technique.

The paper presents MIMO field performance results observed using a ultra mobile broadband (UMB) testbed network. We evaluate metrics such as antenna correlations and channel condition number to characterize the MIMO channel. Results show that low condition numbers, which are beneficial to MIMO, are prevalent for a majority of the coverage area in our network. We demonstrate that the use of MIMO provides gains of the order of 20-40% over SIMO transmissions. These gains are made possible by the use of cross-polarized transmit antennas and advanced UMB features that allow dynamic MIMO vs. SIMO transmission selection based on channel conditions. These results are obtained in a truly mobile, wireless wide-area deployment, which makes them unique. Our results point to the viability and value of MIMO in future mobile wireless networks.