Extracting Sentence Embeddings from Pretrained Transformer Models

Background/introduction: Pre-trained transformer models shine in many natural language processing tasks and therefore are expected to bear the representation of the input sentence or text meaning. These sentence-level embeddings are also important in retrieval-augmented generation. But do commonly used plain averaging or prompt templates surface it enough? Methods: Given 110M parameters BERT's hidden representations from multiple layers and multiple tokens we tried various ways to extract optimal sentence representations. We tested various token aggregation and representation post-processing techniques. We also tested multiple ways of using a general Wikitext dataset to complement BERTs sentence representations. All methods were tested on 8 Semantic Textual Similarity (STS), 6 short text clustering, and 12 classification tasks. We also evaluated our representation-shaping techniques on other static models, including random token representations. Results: Proposed representation extraction methods improved the performance on STS and clustering tasks for all models considered. Very high improvements for static token-based models, especially random embeddings for STS tasks almost reach the performance of BERT-derived representations. Conclusions: Our work shows that for multiple tasks simple baselines with representation shaping techniques reach or even outperform more complex BERT-based models or are able to contribute to their performance.

Inference acceleration for large language models using "stairs" assisted greedy generation

Large Language Models (LLMs) with billions of parameters are known for their impressive predicting capabilities but require lots of resources to run. With their massive rise in popularity, even a small reduction in required resources could have an impact on environment. On the other hand, smaller models require fewer resources but may sacrifice accuracy. In this work, we are proposing an implementation of ``stairs'' assisted greedy generation. It is a modified assisted generation methodology that makes use of a smaller model's fast generation, large model's batch prediction, and "stairs" validation in order to achieve a speed up in prediction generation. Results show between 9.58 and 17.24 percent inference time reduction compared to a stand-alone large LLM prediction in a text generation task without a loss in accuracy.

Sentiment Analysis of Lithuanian Online Reviews Using Large Language Models

Sentiment analysis is a widely researched area within Natural Language Processing (NLP), attracting significant interest due to the advent of automated solutions. Despite this, the task remains challenging because of the inherent complexity of languages and the subjective nature of sentiments. It is even more challenging for less-studied and less-resourced languages such as Lithuanian. Our review of existing Lithuanian NLP research reveals that traditional machine learning methods and classification algorithms have limited effectiveness for the task. In this work, we address sentiment analysis of Lithuanian five-star-based online reviews from multiple domains that we collect and clean. We apply transformer models to this task for the first time, exploring the capabilities of pre-trained multilingual Large Language Models (LLMs), specifically focusing on fine-tuning BERT and T5 models. Given the inherent difficulty of the task, the fine-tuned models perform quite well, especially when the sentiments themselves are less ambiguous: 80.74% and 89.61% testing recognition accuracy of the most popular one- and five-star reviews respectively. They significantly outperform current commercial state-of-the-art general-purpose LLM GPT-4. We openly share our fine-tuned LLMs online.

Optimizing Large Language Models for OpenAPI Code Completion

Recent advancements in Large Language Models (LLMs) and their utilization in code generation tasks have significantly reshaped the field of software development. Despite the remarkable efficacy of code completion solutions in mainstream programming languages, their performance lags when applied to less ubiquitous formats such as OpenAPI definitions. This study evaluates the OpenAPI completion performance of GitHub Copilot, a prevalent commercial code completion tool, and proposes a set of task-specific optimizations leveraging Meta's open-source model Code Llama. A semantics-aware OpenAPI completion benchmark proposed in this research is used to perform a series of experiments through which the impact of various prompt-engineering and fine-tuning techniques on the Code Llama model's performance is analyzed. The fine-tuned Code Llama model reaches a peak correctness improvement of 55.2% over GitHub Copilot despite utilizing 25 times fewer parameters than the commercial solution's underlying Codex model. Additionally, this research proposes an enhancement to a widely used code infilling training technique, addressing the issue of underperformance when the model is prompted with context sizes smaller than those used during training.

Time-Adaptive Recurrent Neural Networks

Data are often sampled irregularly in time. Dealing with this using Recurrent Neural Networks (RNNs) traditionally involved ignoring the fact, feeding the time differences as additional inputs, or resampling the data. All these methods have their shortcomings. We propose an elegant alternative approach where instead the RNN is in effect resampled in time to match the time of the data. We use Echo State Network (ESN) and Gated Recurrent Unit (GRU) as the basis for our solution. Such RNNs can be seen as discretizations of continuous-time dynamical systems, which gives a solid theoretical ground for our approach. Similar recent observations have been made in feed-forward neural networks as neural ordinary differential equations. Our Time-Adaptive ESN (TAESN) and GRU (TAGRU) models allow for a direct model time setting and require no additional training, parameter tuning, or computation compared to the regular counterparts, thus retaining their original efficiency. We confirm empirically that our models can effectively compensate for the time-non-uniformity of the data and demonstrate that they compare favorably to data resampling, classical RNN methods, and alternative RNN models proposed to deal with time irregularities on several real-world nonuniform-time datasets.

Towards Lithuanian grammatical error correction

Everyone wants to write beautiful and correct text, yet the lack of language skills, experience, or hasty typing can result in errors. By employing the recent advances in transformer architectures, we construct a grammatical error correction model for Lithuanian, the language rich in archaic features. We compare subword and byte-level approaches and share our best trained model, achieving F$_{0.5}$=0.92, and accompanying code, in an online open-source repository.

Correcting diacritics and typos with ByT5 transformer model

Due to the fast pace of life and online communications, the prevalence of English and the QWERTY keyboard, people tend to forgo using diacritics, make typographical errors (typos) when typing. Restoring diacritics and correcting spelling is important for proper language use and disambiguation of texts for both humans and downstream algorithms. However, both of these problems are typically addressed separately, i.e., state-of-the-art diacritics restoration methods do not tolerate other typos. In this work, we tackle both problems at once by employing newly-developed ByT5 byte-level transformer models. Our simultaneous diacritics restoration and typos correction approach demonstrates near state-of-the-art performance in 13 languages, reaching >96% of the alpha-word accuracy. We also perform diacritics restoration alone on 12 benchmark datasets with the additional one for the Lithuanian language. The experimental investigation proves that our approach is able to achieve comparable results (>98%) to previously reported despite being trained on fewer data. Our approach is also able to restore diacritics in words not seen during training with >76% accuracy. We also show the accuracies to further improve with longer training. All this shows a great real-world application potential of our suggested methods to more data, languages, and error classes.

Automated age-related macular degeneration area estimation -- first results

This work aims to research an automatic method for detecting Age-related Macular Degeneration (AMD) lesions in RGB eye fundus images. For this, we align invasively obtained eye fundus contrast images (the "golden standard" diagnostic) to the RGB ones and use them to hand-annotate the lesions. This is done using our custom-made tool. Using the data, we train and test five different convolutional neural networks: a custom one to classify healthy and AMD-affected eye fundi, and four well-known networks: ResNet50, ResNet101, MobileNetV3, and UNet to segment (localize) the AMD lesions in the affected eye fundus images. We achieve 93.55% accuracy or 69.71% Dice index as the preliminary best results in segmentation with MobileNetV3.

Generating abstractive summaries of Lithuanian news articles using a transformer model

In this work, we train the first monolingual Lithuanian transformer model on a relatively large corpus of Lithuanian news articles and compare various output decoding algorithms for abstractive news summarization. Generated summaries are coherent and look impressive at the first glance. However, some of them contain misleading information that is not so easy to spot. We describe all the technical details and share our trained model and accompanying code in an online open-source repository, as well as some characteristic samples of the generated summaries.

Efficient implementations of echo state network cross-validation

Background/introduction: Cross-validation is still uncommon in time series modeling. Echo State Networks (ESNs), as a prime example of Reservoir Computing (RC) models, are known for their fast and precise one-shot learning, that often benefit from good hyper-parameter tuning. This makes them ideal to change the status quo. Methods: We suggest several schemes for cross-validating ESNs and introduce an efficient algorithm for implementing them. This algorithm is presented as two levels of optimizations of doing $k$-fold cross-validation. Training an RC model typically consists of two stages: (i) running the reservoir with the data and (ii) computing the optimal readouts. The first level of our proposed optimization addresses the most computationally expensive part (i) and makes it remain constant irrespective of $k$. It dramatically reduces reservoir computations in any type of RC system and is enough if $k$ is small. The second level of optimization also makes the (ii) part remain constant irrespective of large $k$, as long as the dimension of the output is low. We discuss when the proposed validation schemes for ESNs could be beneficial, three options for producing the final model and empirically investigate them on six different real-world datasets, as well as do empirical computation time experiments. We provide the code in an online repository. Results: Proposed cross-validation schemes give better and more stable test performance in all the six different real-world datasets, three task types. Empirical run times confirm our complexity analysis. Conclusions: In most situations $k$-fold cross-validation of ESNs and many other RC models can be done for virtually the same time complexity as a simple single-split validation. Space complexity can also remain the same in all the cases. This enables cross-validation to become a standard practice in reservoir computing.