Andrey Simanovsky

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Vladislav Lialin, Rahul Goel, Andrey Simanovsky et al.

Currently used semantic parsing systems deployed in voice assistants can require weeks to train. Datasets for these models often receive small and frequent updates, data patches. Each patch requires training a new model. To reduce training time, one can fine-tune the previously trained model on each patch, but naive fine-tuning exhibits catastrophic forgetting - degradation of the model performance on the data not represented in the data patch. In this work, we propose a simple method that alleviates catastrophic forgetting and show that it is possible to match the performance of a model trained from scratch in less than 10% of a time via fine-tuning. The key to achieving this is supersampling and EWC regularization. We demonstrate the effectiveness of our method on multiple splits of the Facebook TOP and SNIPS datasets.

Alexander Ulanov, Andrey Simanovsky, Manish Marwah

Present day machine learning is computationally intensive and processes large amounts of data. It is implemented in a distributed fashion in order to address these scalability issues. The work is parallelized across a number of computing nodes. It is usually hard to estimate in advance how many nodes to use for a particular workload. We propose a simple framework for estimating the scalability of distributed machine learning algorithms. We measure the scalability by means of the speedup an algorithm achieves with more nodes. We propose time complexity models for gradient descent and graphical model inference. We validate our models with experiments on deep learning training and belief propagation. This framework was used to study the scalability of machine learning algorithms in Apache Spark.