With the advent of large language models (LLMs), there has been a great deal of interest in applying them to solve difficult programming tasks. Recent work has demonstrated their potential at program optimization, a key challenge in programming languages research. We propose a blackbox adaptation method called Retrieval Augmented Search (RAS) that performs beam search over candidate optimizations; at each step, it retrieves in-context examples from a given training dataset of slow-fast program pairs to guide the LLM. Critically, we find that performing contextual retrieval based on an LLM-generated natural language description significantly outperforms retrieval based on the source code. In addition, we propose a method called AEGIS for improving interpretability by decomposing training examples into "atomic edits" that are significantly more incremental in nature. We show that RAS performs 1.8$\times$ better than prior state-of-the-art blackbox adaptation strategies, and that AEGIS performs 1.37$\times$ better while performing significantly smaller edits.

We evaluate the use of software interpretation to push High Level Synthesis of application-specific accelerators toward a higher level of abstraction. Our methodology is supported by a formal power consumption model that computes the power consumption of accelerator components, accurately predicting the power consumption on new designs from prior optimization estimations. We demonstrate how our approach simplifies the re-use of power optimizations across distinct designs, by leveraging the higher level of design abstraction, using two accelerators representative of the robotics domain, implemented through the Bambu High Level Synthesis tool. Results support the research hypothesis, achieving predictions accurate within +/- 1%.

CP decomposition is a powerful tool for data science, especially gene analysis, deep learning, and quantum computation. However, the application of tensor decomposition is largely hindered by the exponential increment of the computational complexity and storage consumption with the size of tensors. While the data in our real world is usually presented as trillion- or even exascale-scale tensors, existing work can only support billion-scale scale tensors. In our work, we propose the Exascale-Tensor to mitigate the significant gap. Specifically, we propose a compression-based tensor decomposition framework, namely the exascale-tensor, to support exascale tensor decomposition. Then, we carefully analyze the inherent parallelism and propose a bag of strategies to improve computational efficiency. Last, we conduct experiments to decompose tensors ranging from million-scale to trillion-scale for evaluation. Compared to the baselines, the exascale-tensor supports 8,000x larger tensors and a speedup up to 6.95x. We also apply our method to two real-world applications, including gene analysis and tensor layer neural networks, of which the numeric results demonstrate the scalability and effectiveness of our method.

Bag of Words (BOW) [1954]: count the occurrences of each word in the documents and use them as features. TF-IDF [1972]: the BOW scores are modified so that rare words have high scores and common words have low scores. Word2Vec [2013]: each word is mapped to a high-dimensional vector called word embedding, which captures its semantic. Word embeddings are learned by a neural network looking for word correlations on a large corpus. RNN [1986]: RNNs compute document embeddings leveraging word context in sentences, which was not possible with word embeddings alone.

OctoML Inc. today introduced a new release of its artificial intelligence platform that includes a collection of highly efficient neural networks. The neural networks are optimized versions of popular open-source AI models that OctoML has fine-tuned. According to the startup, the optimized versions cost less to run than the original AI models and require less power as well. The new platform release also introduces other improvements, including support for more machine learning development tools. Seattle-based OctoML launched in 2019 and is backed by more than $130 million in funding.

Google today announced the preview launch of TensorBoard.dev "You'll now be able to host and track your ML experiments and share them publicly, no setup required. Simply upload your logs and share the URL so that others can see the experiments and what you're doing with TensorBoard," Google VP of engineering Megan Kacholia said onstage today at TensorFlow World in Santa Clara, California. TensorFlow Enterprise is made to deliver an optimized version of its open source machine learning framework TensorFlow for large businesses. It works with Google's AI Platform and Kubernetes Engine as well as optimized versions of Deep Learning VMs and Deep Learning Containers. The service is made to supply up to 3x improvements in data reading -- the result of changes to how TensorFlow reads and caches files -- and up to 3 years of support for security patches and select bug fixes.

In our webinar "Optimizing Machine Learning with TensorFlow" we gave an overview of some of the impressive optimizations Intel has made to TensorFlow when using their hardware. You can find a link to the archived video here. During the webinar, Mohammad Ashraf Bhuiyan, Senior Software Engineer in Intel's Artificial Intelligence Group, and myself spoke about some of the common use cases that require optimization as well as benchmarks demonstrating order-of-magnitude speed improvements when running on Intel hardware. TensorFlow, Google's library for machine learning (ML), has become the most popular machine learning library in a fast-growing ecosystem. This library has over 77k stars on GitHub and is widely used in a growing number of business critical applications.