LSTM and the Sauna: Connecting RNN, Time-Series, and Thoughts

I take saunas and love the heat. In an old part of the barn, adjacent to my processing computer and my clandestine maps (which were almost discarded around the year 2000), I built my version of a hot box LSTM in an odd little environmental sampling shelf corner.

Long Short-Term Memory (LSTM), defined in various ways, serves here, for my purposes, as a simplified representation of a recurrent neural network capable of remembering information for extended periods. In this box, I release the floodgates of my time-series data files, allowing me to remember or forget the day's little oddities, past moments realized and recorded but not fully dissected for further analysis. It is an effective, efficient, and novel method for me to maintain information over the extended period of my 42 years on planet Earth.

Sometimes, I am taken away by a thought so deep and tucked back into the recesses of my brain that it surprises me. In this place, stressed to the limit at 180, 185, 195 degrees, and higher, surrounded by dry redwood, I am transported back to time immemorial. I find myself in a cave—a dark, hot cave somewhere in time, like a long-lost ancestor, a ghost of the human experience. Stories, ideas, great conceptions, and misconceptions emerge, moments of Zen and pain wrapped into a euphoric cocktail, as the heat gives way to the cold water shower just outside.

This is me, as I think about coding, the next great scientific discovery, pondering why Van Gogh was underappreciated in his day. Through this method, I do what an LSTM application might attempt. I too collect and preprocess data, build a model from it, train it on experience, and develop it into a form where it can be further evaluated, with final predictions and conclusions drawn. Sometimes, I have to run thoughts through an analogy to normalize the data, making it accessible. At times, the day's experiences need to be windowed, similar to seismic data, so that overlapping data points can be better understood as they touch, or non-overlapped so they are not mistakenly correlated as one and the same thought.

So, while an LSTM might be a technical, computational model and the Sauna Experience a deeply human, subjective one, there are parallels in complexity, the process of retaining and discarding information, adaptive responses, layered processing, the importance of temporal aspects, and the influence of the environment. Both LSTM and the Sauna involve intricate processes of filtering and focusing on certain elements while letting others recede; it’s all about what is taken in, what is let go, and how we mesh the data.