LLMs require massive compute for training, posing infrastructure challenges to broader ML community eager to fine-tune models. This poster explains how open-source orchestrators like Flyte provide declarative infrastructure specification to streamline complex LLM workflows, utilize techniques like RAG and 8-bit quantization for model freshness and efficiency, optimize resource usage, and ultimately unlock the full potential of LLMs by making them more accessible.
Unlike traditional Japanese haiku, modern English haiku doesn’t entirely adhere to a 5-7-5 syllable count. A haiku is expected to connect nature with human experience. In contrast, a senryu is expected to focus solely on human feelings and is mainly expected to have humor or satire, often self-deprecating. Classifying poems as haiku or senryu is challenging as such creative works are subjective and open to multiple interpretations.
The work presented in this poster aims to classify haiku and senryu using text classification techniques such as topic identification ( whether or not the poem is focused on nature or purely human feeling), sentiment analysis ( whether the poem is serious or humorous), and named entity recognition (NER) to identify kigo or non-kigo words.
Vision is a screening and data processing platform for cognitive impairment monitoring. Vision helps HCP's(hospital care plan) screen for changes in cognitive impairment as aging patients routinely get tested over time. The Vision testing process is two-fold, Part A is an MMSE(Mini Mental State Examination) test created using a WordPress website along with a form, where HCP's can enter patient test results which can then be related to an estimated level of cognitive impairment. Part B involves an EEG(electroencephalogram) test which has become more and more established within academia as a screening method for Alzheimer. The EEG data is gathered using an OpenBCI and is then processed by a signal processing algorithm on Google Colab, and can then be compared to literature values within.
Multilabel classification is a machine learning task in which each instance is assigned to a group of labels. It has gained widespread use in various applications in recent years. Preprocessing, such as feature selection, is an important step in any machine learning or data mining task. A new algorithm called Black Hole, inspired by the phenomenon of black holes, has recently been developed to tackle multi-label classification problems.
Learn about a modified version of the Black Hole algorithm that combines it with two genetic algorithm operators: crossover and mutation in this poster. The combination of Black Hole and genetic algorithms has the potential to solve multi-label classification problems across a range of domains.
In retail, what we put on the shelves and how we manage stock can really make or break sales. This poster session dives into how Python can tackle these challenges through market basket analysis and utility itemset mining. We'll explore Python techniques for analyzing transactional data to optimize product placement and inventory levels. Attendees will learn to transform retail data into strategic insights, enhancing product mix and boosting sales.
Following the closing of the global NPO Women Who Code, we would like to have a poster about the relaunch of our community, Women in Software Engineering Japan (formerly Women Who Code Tokyo), and make it known that we are still supporting women in software engineering here in Japan.
