If you've been following along with my research, you may have noticed that I haven't posted in a while - three days shy of two months to be exact. There is a number of reasons for this. The primary reason being because I've been struggling and progress has been slow. I haven't felt like there has been much to write about nor have I really wanted to write. Before we get to that though, let's celebrate some wins!

After my last post, I spent some further time creating a SpeechBrain direct SLU recipe for the STOP dataset and successfully trained a model. 🙌🏻

On the surface, that might not seem like much. What it means as we dig a little deeper is that I successfully set up AWS, configured SageMaker, installed SpeechBrain and its dependencies (there were issues with the SageMaker base image), learned about SpeechBrain recipes and HyperPyYAML, understood the STOP dataset, and figured out how to apply the existing direct SLU recipe to a new dataset, which included understanding how to retrain the tokenizer. Now that's a win.

Beyond that, I've done a lot of reading. I've been exploring SLU as a seq2seq problem (versus classification) and therefore learning about encoder-decoder architectures, RNNs, LSTMs, GRUs... I also needed to gain a better understanding of the attention mechanism, to better understand the transformer architecture, to then go on to learn about conformer and branchformer architectures. More recently, I've been reading about Mamba, a new sequence modelling architecture that doesn't rely on transformers and instead incorporates selective state spaces. Although reading these papers is interesting, I haven't been giving them my full attention. All along there has been a nagging thought at the back of my mind: why?

Why am I doing a masters degree? Right now, I'm not too sure. You see, there was a time when I wanted to do my Ph.D., and for that the masters made sense, but that's no longer the case. The next goal was to land a machine learning job. I checked this off long before completing the masters, albeit as a solutions engineer in pre-sales and not a full-on machine learning engineer. That said, through working at Edge Impulse, I discovered developer relations and recognized that the role is a much better fit for my skills and personality than the typical machine learning engineer position. In fact, I'll be switching over to DevRel at Edge Impulse next month. So I start to question if the masters is even needed. At some point my experience in the field will outweigh the benefits of the degree.

I've wanted to quit more times than I care to admit. I'm in that shitty phase of grad school where there is a lot of unknown. As my colleague put it, "I'm wandering in the wilderness." What I will add is that it feels like I'm wandering in a dense rainforest, at night, with no torch, and I just can't seem to find the path, all while my mind is dealing with doubtful thoughts. Will I ever finish this friggen thing? Is the effort worth it? What's the cost? What's the payoff? Is the topic relevant? Is my research novel enough? Does this align with what I want out of life?

For years I've had this vision for my life. It goes back as far as high school and the seed was planted even earlier, in elementary school. These days I call it, "The GoPro Life." A life full of adventure, exploration, community, and joy. It has had different flavours over time yet it always seems to be centered around adventure travel content creation. Think telling the stories of adventurers and participating in their passions with them, documenting outdoor events and festivals, bringing people on surprise trips, highlighting unique parts of the world, and the like. At the same time, I want to figure out how to bring in the tech/engineering side of me as well. Think working with scientists to deploy edge AI solutions in the Caribbean for Shark Week (and documenting it of course). I just haven't figured out how to tie it all together and make a living too.

Coming back to the masters, I spend a lot of time contemplating if the effort is worth it, is the juice worth the squeeze as they say. A masters isn't critical to my vision and I wonder if my time is better spent finishing the masters or working on that dream? When I break it down, I know what I need to do. I need to stop questioning and doubting. I need to flip that energy towards certainty, doing the work, graduating, and moving on. It's about who I become in the process, not the letters after my name. Besides, having the degree is not going to hurt.

Going through the masters process thus far has already brought me benefits. It likely helped me get my current job, what I've been learning has come up in conversations, my confidence in machine learning has grown, it was the impetus for this blog, it helped me clarify my vision for my life, and more. Results certainly worth a celebration dance. Yet contradictorily, I find the masters work out of alignment, draining, and as if it's pulling me away from where I want to go. Quite the juxtaposition and hence the title of this post: dancing on quicksand.