Srikanth Cherla

Migrating to Hugo

Thu, 28 May 2026 00:00:00 +0000

This website has been running on WordPress since I first set it up in 2013 — hosted on a shared cPanel server at Arvixe (and then Domain Racer). It has served me well, but over time the maintenance overhead started to feel disproportionate to what I actually needed: a simple personal site with a blog, a CV, a music section, and not much else. WordPress is a powerful platform, but for a use case like mine it brings a lot of complexity along with it — a database, PHP, plugins to keep updated, and a web-based admin interface that I rarely enjoyed using.

New Guitar Video – Just the Same by Gentle Giant

Sun, 14 Sep 2025 00:00:00 +0000

Just the Same opens Free Hand (1972), my favourite among all of Gentle Giant’s brilliant albums. The song is built around interlocking guitar and keyboard figures with the characteristic rhythmic unpredictability that defines their writing — never settling long enough to feel comfortable, which is exactly what makes it so interesting to play.

New Guitar Video – Corpse Pose by Unwound

Sat, 16 Aug 2025 00:00:00 +0000

Corpse Pose is from Leaves Turn Inside You (2001), the final record Unwound made before disbanding. It’s a quieter, more abstract piece than much of their earlier work — part of an album that saw them pushing further into noise and texture. One of those songs where the space between notes matters as much as the notes themselves.

New Guitar Video – Revolve by Melvins

Mon, 11 Aug 2025 00:00:00 +0000

Revolve is from Houdini (1993), one of the Melvins’ most accessible albums. It’s also the first Melvins song I ever heard, which got me curious about the band (that verse riff!). Melvins is one of those bands that you don’t really get a lot of their stuff right away, but the longer you stick to it the more rewarding it gets. This song is just an easily palatable teaser to the mammoth collection of songs they’ve written. Having seen them live a couple of times, and being a huge fan, I’ve been wanting to learn their music for a while.

New Guitar Video – You Think I Ain't Worth a Dollar by Queens of the Stone Age

Wed, 16 Jul 2025 00:00:00 +0000

The opening track from Songs for the Deaf (2002), one of QOTSA’s finest records. It’s a fast, aggressive sprint of a song — under two minutes — that sets the tone for the whole album perfectly. The guitar tone on the original recording is enormous. Fun to play loud.

Completed Generative AI with LLMs Course on Coursera

Thu, 01 Aug 2024 00:00:00 +0000

I recently completed a Coursera foundation course on Large Language Models — my first structured learning in a while. The gap was mostly down to becoming a parent, which makes carving out time for professional development and blogging considerably harder.

Since April 2023 I’ve been working on Muse at Unity — an LLM-driven AI assistant for Unity developers that has evolved from a web-based chat interface into a deeply integrated Editor tool capable of analysing your project and performing contextual tasks.

Unity Muse at the Microsoft Build Conference

Sat, 25 May 2024 00:00:00 +0000

I’ve been building Unity’s AI assistant — Muse — since April 2023, and it recently got a moment in the spotlight: Microsoft featured it at their Build conference as a customer success story.

A film crew came to Copenhagen to record the team, and the resulting video was shown during a live Muse demonstration at the conference. I’m in it, talking about both the product and my contributions to it. A good reminder of how far the project has come from its early days as a simple chat interface.

New Guitar Video – Invincible by Tool

Sat, 18 May 2024 00:00:00 +0000

Tool is one of my favourite bands, and Invincible is a piece I’d been wanting to cover for a while. Practice time has shrunk considerably since becoming a parent, but I made this one a priority.

The challenge wasn’t technical complexity — it was sustaining focus across a very long piece without losing your place. I usually aim for single takes on all my videos. This one needed two: I slipped near the end of the first attempt.

Completed Learn TypeScript Course on Codecademy

Tue, 03 May 2022 00:00:00 +0000

I returned from paternity leave to find the team had shifted from Unity’s C# codebase to TypeScript. Having just finished the C# course on Codecademy in December, I was glad to find they had a TypeScript course too.

Prior experience with Python’s mypy type-checker made TypeScript’s type system feel familiar — the conceptual leap was smaller than expected. I also found useful parallels in package management and scripting approaches between languages.

Completed Learn C# Course on Codecademy

Sat, 08 Jan 2022 00:00:00 +0000

I’d been at Unity for over a year without ever properly learning C#. Advanced Unity tutorials weren’t the right entry point — too much assumed knowledge. During a month-long trip to India in December 2021 I went back to basics with Codecademy’s beginner C# course.

The browser-based editor wasn’t enough for me. I set up a local environment with Mono, Dotnet, Omnisharp, and Vim, and made sure to type every piece of code myself — even the repetitive parts like imports and base classes. Repetition was the key.

I'm a Dad!

Tue, 28 Dec 2021 00:00:00 +0000

New Guitar Video – Man in the Box (Acoustic) by Alice in Chains

Mon, 23 Aug 2021 00:00:00 +0000

An original acoustic arrangement of Man in the Box by Alice in Chains — released as a tribute on the 30th anniversary of their debut album Facelift. Video editing by my wife, Nina.

Completed Practical Reinforcement Learning on Coursera

Sat, 01 May 2021 00:00:00 +0000

Reinforcement Learning is one of the most interesting topics in computer science and ML. I started with the canonical textbook — Reinforcement Learning: An Introduction by Sutton & Barto — which is a fantastic read, though some topics took multiple passes to properly absorb.

After working through Dynamic Programming, Monte-Carlo methods, and Temporal Difference learning, I wanted practical experience and turned to Coursera’s Practical Reinforcement Learning course.

Honest review: it was frustrating. The course tried to cover too much ground too quickly, assignment instructions were minimal and error feedback sparse, and by the halfway point I was no longer enjoying it. I pushed through to the end.

New Guitar Video – Rusty Cage by Soundgarden

Sat, 06 Mar 2021 00:00:00 +0000

My favourite track from the Road Rash video game soundtrack — see the Outshined post for how I came across both songs.

I skipped the longish intro because aligning it with the backing track without metronome cues proved too awkward. Practised alongside the Outshined cover and recorded both around the same time.

New Guitar Video – Outshined by Soundgarden

Wed, 24 Feb 2021 00:00:00 +0000

My connection to this song goes back to before I even knew what Soundgarden or grunge music were. I first heard it on the Road Rash video game soundtrack — a game that introduced me to Therapy?, Monster Magnet, and two Soundgarden tracks: this one and Rusty Cage.

A fond look back at those days, with the tuning dropped a step for a more satisfying effect.

New Guitar Video – Superunknown by Soundgarden

Mon, 25 Jan 2021 00:00:00 +0000

Despite Soundgarden being one of my favourite bands from the 1990s grunge era, I’d never learned to play any of their songs. This cover of Superunknown — the title track from their landmark album — is hopefully the first of a few I’ll upload.

Performed on a PRS guitar in drop tuning to match the backing track.

New Guitar Video – Americana by the Offspring

Thu, 24 Dec 2020 00:00:00 +0000

A cover of Americana by the Offspring — not one of the album’s most popular tracks, but one of the most fun and energetic ones to play. The album has special personal significance from when I first bought it as a teenager.

My last video of 2020. Happy to be producing more next year. Happy Christmas!

Completed Game Design Foundations on LinkedIn Learning

Sun, 22 Nov 2020 00:00:00 +0000

After joining Unity I needed to get up to speed on game design. I’d moved away from gaming as a consumer over the years to focus on ML and music technology, so coming in fresh to the industry I wanted a solid foundation.

I found a three-part course called Game Design Foundations taught by Brenda Romero on LinkedIn Learning and completed it over about two and a half weeks. It gave me the vocabulary — terms like “core loop” and “lootbox” that colleagues were using — and helped me start to see how my ML background could contribute to game design problems.

Hello, Unity!

Mon, 02 Nov 2020 00:00:00 +0000

I’m very pleased to announce that, starting today, I will be a Senior Machine Learning Developer at Unity Labs in Copenhagen. Its parent company - Unity Technologies, is well-known for having produced one of the most widely used gaming engines - Unity. I’m super-excited about this change of focus in my work from music to gaming, and really look forward to getting started!

At Unity Labs, I will be creating Machine Learning solutions in technology for use by Game Designers. This is about all I know for now, and hope that I can share more updates as time passes.

New Guitar Video – Wynona's Big Brown Beaver by Primus

Mon, 21 Sep 2020 00:00:00 +0000

Third in my series of Primus covers. This one was quite challenging, especially the second solo — a good showcase of Larry Lalonde’s unique and quirky playing style. I’ll be exploring other artists before returning for more Primus.

Goodbye, Moodagent!

Tue, 08 Sep 2020 00:00:00 +0000

Today I handed in my resignation at Moodagent. It’s been a great year and some months working in this fast and ambitious company! I will associate this experience most with the great friends I’ve made here, my focus on data preparation, Apache Spark, Collaborative Filtering and a feeling that I’ve really improved my programming abilities thanks to some excellent Coursera courses (this one, and this one) I completed while working here.

New Guitar Video – Jerry was a Racecar Driver by Primus

Sat, 13 Jun 2020 00:00:00 +0000

Second in my planned series of five Primus covers. Jerry was a Racecar Driver is one of the band’s more recognisable pieces — a nice and easy exercise in volume swells, and a crazy solo that jumps in and out of scale alongside one of Les Claypool’s most intricate basslines.

New Guitar Video – John the Fisherman by Primus

Thu, 11 Jun 2020 00:00:00 +0000

I’ve long admired Larry Lalonde’s distinctive guitar work alongside Les Claypool’s intricate bass playing. To properly appreciate it I set out to learn five Primus songs.

John the Fisherman is the first — not a difficult song, really, and a lot of fun to play. A great one to get started with. Also the inaugural video on my new Ibanez RG-3120.

Completed Programming Languages (Part B) on Coursera

Sat, 06 Jun 2020 00:00:00 +0000

Completed the second of three parts of Dan Grossman’s Programming Languages course from the University of Washington on Coursera. Where Part A used Standard ML, Part B uses Racket — exploring dynamically typed languages and culminating in implementing a simple language interpreter.

Part A gave me a solid foundation in functional programming and formalised a lot of intuitions I’d built up informally. I’ll share fuller reflections once I’ve finished Part C and had a chance to look back at the whole series.

New Guitar Video – Perpetual Black Second by Meshuggah

Thu, 23 Apr 2020 00:00:00 +0000

Perpetual Black Second is from Nothing (2002), recorded during the first weeks of pandemic lockdown. Meshuggah’s guitar playing sits in a category of its own — the polyrhythmic layering creates a hypnotic, disorienting effect that I’ve found fascinating since first hearing Destroy Erase Improve. Their guitars are tuned down to F, which accounts for a significant part of that heaviness.

New Guitar Video – The Pot by Tool

Thu, 26 Mar 2020 00:00:00 +0000

I wasn’t immediately taken with this song but reconnected with it while preparing for Tool’s Fear Inoculum release, and got drawn in enough to learn it properly.

My first video on a new PRS SE Mark Holcomb Signature Edition. Played in Drop C standard.

Completed Programming Languages (Part A) on Coursera

Mon, 09 Mar 2020 00:00:00 +0000

Completed the first part of Dan Grossman’s Programming Languages specialisation from the University of Washington on Coursera, scoring 98%. The course uses Standard ML to teach functional programming — pattern matching, function closures, partial application, currying, mutual recursion.

Beyond the technical content, what stood out was the emphasis on programming style and code quality as part of the evaluation criteria. The assignments were rigorous and the handouts were thorough. An excellently designed course, even if challenging.

Attending ISMIR 2019

Tue, 08 Oct 2019 00:00:00 +0000

Four of us from Moodagent — Reinier de Valk, Pierre Lafitte, Tomas Gajarsky, and myself — are attending ISMIR 2019 in Delft, Netherlands.

Two presentations from the team:

Reinier de Valk: “JosquIntab: A Dataset for Content-based Computational Analysis of Music in Lute Tablature” (main conference)
Tomas Gajarsky: “Reinforcement Learning Recommender System for Modelling Listening Sessions” (late-breaking session)

Come find us at the posters if you’re attending.

New Guitar Video – Hangar 18 by Megadeth

Tue, 13 Aug 2019 00:00:00 +0000

Hangar 18 from Rust in Peace (1990) is essentially a vehicle for one of thrash metal’s great dual-guitar solo sections. Marty Friedman and Dave Mustaine trade leads across an extended passage that accounts for most of the song’s final two minutes. A considerable technical challenge, but one of the more satisfying ones.

Passed Rock School Grade 7 (Electric Guitar) Exam… Unofficially

Sat, 06 Jul 2019 00:00:00 +0000

Two years after passing Grade 6 with distinction, I spent about a year preparing for Grade 7. Since I’m now outside the UK, I sat a mock examination via Skype with my guitar tutor Nicolas.

Nicolas scored me at 88 out of 100 — a strong pass, just below distinction. Unofficial, but a definite pass. I’ve uploaded the three exam pieces to YouTube.

Jeff Beck – Cause We’ve Ended as Lovers

Hello, Moodagent!

Thu, 04 Jul 2019 00:00:00 +0000

It’s been about four months since I wrote here about leaving Jukedeck. So after a nice long break, I’m very pleased to share that I’ll be joining Danish music streaming startup Moodagent on the 17th of July, 2019. While the streaming service itself is new and hasn’t been launched yet, the company Moodagent A/S that owns it has been around for nearly two decades having built several products around their core technology for analysing musical content. You may have even come across their first music app on your Nokia phone back in the day! You can read all about them on Wikipedia, and find out more about the Moodagent streaming service on their website. I hear they’ll be launching it very soon!

Getting Started with Python Pandas

Sun, 09 Jun 2019 00:00:00 +0000

I resisted pandas for a long time, then found myself spending way too much time writing data processing routines from scratch on a recommender systems project. It reminded me of when I used to manually compute neural network gradients before discovering Theano.

Two resources that worked well for getting up to speed quickly:

Kaggle Learn micro-courses — well-organised, beginner-friendly, and the pandas course pairs nicely with their visualisation and embedding courses.

A New Blog Series on the Music Tech Community – India Website

Fri, 24 May 2019 00:00:00 +0000

I’ve been volunteering with peers in India to promote awareness of music technology through the Music Tech Community – India initiative. We’re launching a blog series featuring interviews with people working in the field — to showcase what’s possible and offer career inspiration.

The first post features an interview with Ajay Srinivasamurthy, a researcher applying Information Retrieval techniques to Indian classical music. Great start — looking forward to more conversations like this.

Completed Recommender Systems Specialisation on Coursera

Thu, 25 Apr 2019 00:00:00 +0000

After leaving Jukedeck I completed the four-course Recommender Systems specialisation from the University of Minnesota on Coursera:

Introduction to Recommender Systems: Non-personalised and Content-based
Nearest Neighbour Collaborative Filtering
Recommender Systems: Evaluation and Metrics
Matrix Factorisation and Advanced Techniques

Completed in about a month at a leisurely pace. Very well-taught — the coursework used spreadsheet-based implementations to make the algorithms tangible before diving into code. Content-based filtering, item-item and user-user collaborative filtering, matrix factorisation — all well covered.

Goodbye, Jukedeck!

Wed, 13 Mar 2019 00:00:00 +0000

This is just a quick post to let everyone know that I have decided to leave Jukedeck. It’s been a unique and fascinating journey the past three or so years with a flexible and forward-thinking company, and a stimulating work environment. I couldn’t have asked for a more apt transition into employment after my PhD than the one that led me to Jukedeck and I’m really grateful for all that I have learned here, the people I’ve had the opportunity to work with and everything the company has done for me during this period. This also means that I’m no longer going to be living or working in the UK, and my wife Nina and I have some new and exciting plans for the future that I’m really looking forward to.

Remote Talk at Event Organised by Music Tech Community – India

Mon, 07 Jan 2019 00:00:00 +0000

I was invited by Music Tech Community – India to speak at an event in Bengaluru on December 29th, focused on “Machine Learning for Art & Music Generation.” Since Nina and I were on holiday in Mararikulam, Kerala at the time, I delivered it remotely via Skype.

The organisers — Albin Correya, Manaswi Mishra, and Siddharth Bharadwaj — made sure everything ran smoothly. I didn’t want to miss the opportunity, and it worked out well. Other speakers included Harshit Agarwal and two of the organisers.

Oral Presentation at ISMIR 2018

Wed, 10 Oct 2018 00:00:00 +0000

I attended ISMIR 2018 in Paris with colleagues from Jukedeck to present our StructureNet paper. This year’s format gave every accepted paper both an oral and a poster slot — a nice change from the traditional division.

I gave the oral presentation; colleagues Gabriele and Marco handled the poster. The talk was recorded and is on YouTube. Gabriele also wrote up a post on the Jukedeck R&D Team’s Medium page.

A highlight was the jam session organised by Uri Nieto from Pandora — I played bass with fellow attendees on two songs, alongside music ranging from AI-composed pieces to jazz, blues, rock, and heavy metal. I also played guitar for a cover of Blackest Eyes by Porcupine Tree, recorded on the boat cruise on the Seine. Also great to reconnect with people from the Music Informatics Research Group at City and to finally get a photo with both my master’s supervisor Hendrik and my PhD supervisor Tillman.

New Guitar Video – Red Barchetta by Rush

Sat, 30 Jun 2018 00:00:00 +0000

The Rush cover I hinted at after Jacob’s Ladder. Red Barchetta is from Moving Pictures (1981), built around a Neil Peart story about a future where vintage cars have been outlawed. Alex Lifeson’s clean-toned arpeggiated intro is one of my favourite moments in classic rock guitar — deceptively simple, and instantly recognisable.

Paper Accepted at ISMIR 2018

Fri, 25 May 2018 00:00:00 +0000

A paper I co-authored with colleagues from Jukedeck has been accepted at ISMIR 2018 in Paris. Congratulations to Gabriele, Katerina, Matt, Samer, Marco, Ed, and Kevin.

Medeot G., Cherla S., Kosta K., McVicar M., Abdallah S., Selvi M., Newton-Rex E., Webster K. “StructureNet: Inducing Structure in Generated Melodies.” Proc. ISMIR 2018. Paris, France.

TensorFlow Tip: Pretrain and Retrain

Sun, 22 Apr 2018 00:00:00 +0000

I recently ran into a situation where I had to initially train a neural network first on one dataset, save it and then load it up later to train it on a different dataset (or using a different training procedure). I implemented this in Tensorflow and thought I’d share a stripped down version of the script here as it could serve as an instructive example on the use of Tensorflow sessions. Note that this is not necessarily the best way of doing this, and it might indeed be simpler to load the original graph and train that graph itself by making its parameters trainable, or something else like that. The script can be found here. In the first stage of this script (the pre-training stage) there is only a single graph which contains the randomly initialised and trained model. One might as well avoid explicitly defining a graph as Tensorflow’s default graph will be used for this purpose. This model (together with its parameters) is saved to a file and then loaded for the second re-training stage. In this second stage, there are two graphs. The first graph is loaded from the saved file and contains the pre-trained model whose parameters are the ones whose values we wish to assign to those of the second model before training the latter on a different dataset. The parameters of the second model are randomly initialised prior to this assignment step. In order for the assignment to work, I found it necessary to assign parameters across graphs and this could be done by saving the parameters of the first model as numpy tensors and assigning the values of these numpy tensors to the right parameters of the second model.

Completed Machine Learning with Big Data on Coursera

Sat, 17 Mar 2018 00:00:00 +0000

Completed UCSD’s Machine Learning with Big Data on Coursera with a 98.9% mark. The ML theory was introductory — a good refresher on Naive Bayes, Decision Trees, and k-Means, but nothing new. The real value was the hands-on introduction to KNIME and Spark ML applied to real datasets.

Together with the previous course this was more practically focused than the earlier modules in the specialisation, which is what I was after.

Completed Big Data Integration and Processing on Coursera

Sat, 17 Feb 2018 00:00:00 +0000

Completed UCSD’s Big Data Integration and Processing on Coursera with 97.7%. Covered MongoDB for querying JSON data, Pandas for data analysis, and various Spark technologies — Spark SQL, Spark Streaming, Spark MLlib, Spark GraphX. A hands-on exercise involved analysing tweets using both MongoDB and Spark.

The specialisation as a whole is fairly introductory — broad coverage rather than depth — but valuable for building awareness of the ecosystem and knowing where to start when you need to apply these tools.

New Guitar Video – Man in the Box by Alice in Chains

Sat, 13 Jan 2018 00:00:00 +0000

This track was what got me into Alice in Chains. I learned it while recovering from a broken finger — a good excuse to work on wah pedal technique, which this song is perfect for. I used a Vox wah pedal, a bit squeaky but it worked.

The recording isn’t perfect — time constraints during recovery — but I’m happy with it. A Rush cover is next.

See also: the acoustic version recorded in 2021.

The TensorFlow Datasets API for Sequence Data (Code Examples)

Mon, 18 Dec 2017 00:00:00 +0000

When TensorFlow 1.4 was released there were very few fully working examples of the Datasets API for sequence data. Rather than a full tutorial, here are two scripts with explanatory notes.

The GitHub repository contains:

placeholder_vs_iterators.py — Three data input approaches:

Traditional placeholder method
Iterators
Feedable iterators

generator_vs_tfrecord.py — Three methods for iterating through sequence data during training:

Generator function with preprocessing (zero-padding, batching)
Pre-processed data via generator
TFRecord files using SequenceExample Protocol Buffers (the most Datasets API-dependent approach)

References: TF Datasets documentation, Google Developers blog post on the API.

Completed Andrew Ng's Convolutional Neural Networks on Coursera

Sun, 03 Dec 2017 00:00:00 +0000

Completed Andrew Ng’s Convolutional Neural Networks course — the third in his Deep Learning specialisation — with 100%. This was the most genuinely new material for me; I’d only skimmed a couple of papers on CNNs and never properly implemented one.

The course is excellent. Highlights: 1D, 2D, and 3D convolutions explained clearly and in depth; coverage of VGGNet, InceptionNet, and Network-in-Network architectures; applications including object recognition, face recognition, and Neural Style Transfer. The programming assignments were engaging and moderately challenging, and the reading list was valuable.

New Guitar Video – Jacob's Ladder by Rush

Sun, 03 Dec 2017 00:00:00 +0000

I learned this piece while recovering from a bicycle accident. Jacob’s Ladder has some fascinating time signature work — sections in 11/8 and 13/8 — which is a big part of what drew me to it. Alex Lifeson’s guitar writing is exceptional throughout.

Another Rush piece is already in the works.

Completed Big Data Modeling and Management Systems on Coursera

Mon, 27 Nov 2017 00:00:00 +0000

Completed the first course in UCSD’s Big Data specialisation — Big Data Modeling and Management Systems — with a perfect score. Broad coverage of relational databases, big data management systems, and various processing alternatives. The content was somewhat superficial on individual topics but useful for building a mental map of the ecosystem.

The peer-graded capstone (a relational database design for a game) had poorly defined objectives and evaluation criteria — a weak ending to an otherwise reasonable course. Looking forward to the more hands-on exercises promised in later courses in the specialisation.

(Automated) Curriculum Learning

Sat, 18 Nov 2017 00:00:00 +0000

I’ve lately spent some time reading about Curriculum Learning and experimenting with the algorithms described in two of the papers in this domain Bengio, Y., Louradour, J., Collobert, R., & Weston, J. (2009, June). Curriculum learning. In Proceedings of the 26th annual international conference on machine learning (pp. 41-48). ACM. Graves, A., Bellemare, M. G., Menick, J., Munos, R., & Kavukcuoglu, K. (2017). Automated Curriculum Learning for Neural Networks. arXiv preprint arXiv:1704.03003. The first of the above can be considered important given how with empirical results supporting Curriculum Learning, it revived the interest among researchers in this technique. The second is one of the recently proposed approaches for Curriculum Learning that I thought would be interesting to understand in greater depth. I’ve summarised my thoughts on these in a short presentation. I hope to share my code and results not too long from now as well.

Completed Introduction to Big Data on Coursera

Mon, 23 Oct 2017 00:00:00 +0000

Completed UCSD’s Introduction to Big Data on Coursera with 98.9%. The course was light — definitions, history, big data jargon, and very basic principles. The section on the Hadoop ecosystem was new to me and included a hands-on Hadoop assignment which I found worthwhile.

Easy course, but a necessary first step in the specialisation. Looking forward to the remaining courses.

introduction-to-big-data-certificate

Completed Andrew Ng's Improving Deep Neural Networks on Coursera

Sun, 01 Oct 2017 00:00:00 +0000

Completed Andrew Ng’s Improving Deep Neural Networks — the second in his Deep Learning specialisation — with 100%. Much of the material was familiar from my ML background, but several sections were genuinely valuable: the detailed treatment of optimisation techniques (exponential moving averages, Momentum, RMSProp, Adam), batch normalisation, and dropout.

On to the Convolutional Neural Networks course next.

improving-deep-neural-networks-certificate

Invited Talks at IIIT-Bangalore and Robert Bosch

Wed, 13 Sep 2017 00:00:00 +0000

I’m on a break from Jukedeck until September 22nd, visiting Bangalore. Past mentors invited me to present at two organisations.

Today I gave a talk at IIIT-Bangalore covering my PhD work on sequence modelling in music — RBMs and Recurrent RBMs. A similar talk is scheduled at Robert Bosch on September 18th. Slides are available as a PDF if anyone is interested.

Participating in CSMC 2017 Panel Discussion

Sat, 02 Sep 2017 00:00:00 +0000

I’ll be participating in a panel discussion titled “Applying Musical Patterns in Generation” at the 2nd Conference on Computer Simulation of Musical Creativity (CSMC), September 11–13, 2017, in Milton Keynes, UK.

Fellow panellists: Elaine Chew, Roger Dean, Steven Jan, David Meredith, and Tillman Weyde. Organised by Iris Yuping Ren. Looking forward to it.

Completed Andrew Ng's Structuring Machine Learning Projects on Coursera

Mon, 28 Aug 2017 00:00:00 +0000

Completed Andrew Ng’s Structuring Machine Learning Projects with 96.7%. Reasonably familiar material given my background, but a few useful insights.

The lectures on Transfer Learning, Multitask Learning, and End-to-End ML were too brief to be immediately useful — they’d need to be followed up with deeper reading. But the practical advice and real-world scenario exercises were valuable, and I wish there were more of them (perhaps as optional material).

structuring-machine-learning-projects

My Experience in Applying for a Work Visa at the UK Home Office

Mon, 17 Jul 2017 00:00:00 +0000

I submitted my Tier 2 work visa application to the UK Home Office on May 9th, with a guaranteed 8-week processing time. Nearly 10 weeks later, with no decision and my passport still held, I contacted them.

The helpdesk was unhelpful — they offered to forward my details to a colleague and said I’d hear back in 3–5 working days. Holding your passport has real consequences: no travel, no financial transactions requiring ID. I contacted my local MP and filed a formal complaint.

Passed Rock School Grade 6 (Electric Guitar) Exam

Wed, 28 Jun 2017 00:00:00 +0000

I passed the RSL Awards Rock School Grade 6 Electric Guitar exam on June 9th with distinction. I recorded performances of the three pieces — with solos composed by myself — and uploaded them to YouTube.

Favela

Mohair Mountain

That Sounds Like Noise

D'oro

Sat, 03 Jun 2017 00:00:00 +0000

I like using the Pomodoro Technique to remind me to take short breaks while I’m immersed in work. And I thought it would be nice to create a little command-line Pomodoro Timer for myself that will pop up desktop notifications telling me that it’s time to take a break. I call this very simple and minimal Pomodoro timer app D’oro and it can be invoked using a command called doro once it has been installed. You can clone the repository from its Gitlab page . I intend to write an installation script and also create a Debian package for it in the future, but it works and I love using it everyday at work!

Breaking Down the Differentiable Neural Computer

Wed, 24 May 2017 00:00:00 +0000

My PhD work on RNNs for musical sequence prediction got me interested in memory-augmented neural architectures. I spent a couple of weeks working through two key papers:

Neural Turing Machine — Graves et al., Google DeepMind (arXiv)
Differentiable Neural Computer — a more advanced variant published in Nature

I put together a Google Slides presentation with my observations and notes. Feedback welcome — let me know if anything needs correcting.

Reflections on Three Months of Remote Work

Thu, 09 Feb 2017 00:00:00 +0000

From October 2016 to January 2017, Nina and I were in Hyderabad, India — and I worked remotely for Jukedeck throughout. Here’s what I learned.

Setup

Before leaving London, my team lead Kevin and I agreed on IST working hours (11 AM–7 PM, giving five hours of overlap with London), daily standup updates via Slack, and asynchronous participation in planning via Google Docs and Skype.

Standups

We initially had a colleague hold a laptop during in-person standups, but audio quality was poor. We switched to asynchronous video messages via Skype — which had the unexpected benefit of keeping standups brief and concise.

Merry Christmas!

Sun, 25 Dec 2016 00:00:00 +0000

My lovely wife Nina and I recorded a little video where we play a cover of the song “Have Yourself a Merry Little Christmas” by Hugh Martin and Ralph Blane, to wish all our loved ones a merry Christmas. So here’s wishing everyone a merry Chirstmas and a very happy New Year from the both of us!

Let's Encrypt for Free!

Mon, 19 Dec 2016 00:00:00 +0000

This is an account of how I went from no encryption, to almost getting a paid SSL certificate to finally making and installing a free one on my domain. It started with me setting up an ownCloud server on my hosting account to access and sync my data on the cloud after going from Dropbox, to Copy to Mega and finally to pCloud over a span of five or so years. Why ownCloud? Mainly because I have a shared hosting account with Arvixe (an excellent hosting service) with unlimited data storage, and I was curious as to how much of an effort it would be to set up my own cloud storage since hearing about ownCloud a couple of months ago. It turns out that it wasn’t much of an effort after all. I simply contacted the support team at Arvixe who made the ownCloud app available on my cPanel and then it was just a matter of filling a simple online form with little details such as where to store your data, which address to access the ownCloud web interface on, etc. The ownCloud project is fantastic! And from what I’ve seen, it has most (if not all) features that any other company like Dropbox or Mega has to offer. It took me 15 minutes to set things up, install the (Linux) client and sync my cloud storage (a folder on my hosting account) with a local folder. So is that it? Turns out that there’s more. Since now I’m transferring data to and from my domain, it is preferred that the connection to the domain is secure. And the connection can be secured with SSL Encryption. SSL Encryption I won’t go much into TLS/SSL encryption here as there are plenty of resources online that explain it. It would suffice to know that it is a way for a website to secure the connection between itself and a visitor so that any data exchanged between the two is encrypted and not visible to a (potentially malicious) third party that is eavesdropping on the connection. This is necessary to prevent what is known as a man-in-the-middle attack where a hacker intercepts the connection between the website and its visitor and collects the data being transmitted between the two (which may sometimes be confidential, such as credit card numbers, personal identification numbers, etc.) without either the website or its visitor knowing about it. [caption id=“attachment_117” align=“aligncenter” width=“700”] The Chromium Browser address bar when the connection to the page is not secured (note the “http://”).[/caption] There has lately been a growing interest on the web to adopt SSL (or its successor TLS) to secure connections between them and their visitors. Google has even proposed to blacklist websites that don’t adopt the SSL protocol. At a first glance, one can know whether or not a website is secure by keeping an eye out for a green lock next to the address bar, and the fact that it says https://(with the green lock symbol) in the addres bar instead of http://. The s here stands for secure. And if you click on the green lock, it pops up a little window that shows who the site has been secured by. [caption id=“attachment_116” align=“aligncenter” width=“700”] The Chromium Browser address bar when the connection to the page is secured (note the “https://”).[/caption] All this stress on security and privacy is, in my opinion, justified. So given that now I’m transmitting my data between my local machine and my domain I decided it would be a good idea to adopt SSL encryption on my domain. This can be realised by obtaining an SSL certificate from a Certification Authority. SSL Certificates and Certification Authorities In order to obtain an SSL certificate for your domains, you should purchase it from a certification authority (CA) or a reseller who sells it at a cheaper rate sans some extra benefits of support that the CA would be able to offer for a higher price. Some of the most popular CAs around are Symantec, GeoTrust, GlobalSign, DigiCert and GoDaddy. Each of these CAs sells you a certificate for a fixed period of time - typically 1 to 3 years - and offers different packages such as Extended Validation, Wildcard domain certification, etc. For instance, have a look at what Symantec, GeoTrust and GlobalSign have to offer. These are very similar options but priced differently depending on the CA’s credibility (which apparently is a major factor in deciding whom to go with) and what is contained in the option. On the other hand, there are companies that purchase certificates from the CAs in bulk and re-sell them at a cheaper rate. These are websites such as SSL Shopper, or even your own hosting company. I know my hosting company Arvixe re-sells certificates purchased from GlobalSign. Depending on whether you are purchasing your certificate from a re-seller or directly from a CA, the price varies between $17 (the lowest I could find for a RapidSSL certificate from SSL Shopper) to a few thousand dollars. A CA or a re-seller issues you a certificate following a verification procedure that confirms that you are indeed the owner of the domain and that your company is a legitimate one whose credentials have been verified by this issuing authority. And the verification process is either manual or fully automatic and depending on how thoroughly it is done, the issuance of a certificate can take anything between a few minutes to a few weeks. I did not complete this process myself (for reasons explained below) but I do recall abandoning a few applications midway because it seemed like a hassle to provide them with information I didn’t even know the meaning of. And although an expensive, time-consuming and thorough process might make sense for a big company that is dealing in a lot of financial transactions and exchange of information with its customers with a lot at stake, I felt like it was an overkill in my case when all I wanted to secure was my personal domain and communications with my ownCloud server (remember?). Now this all sounds good, and I was almost convinced that I should buy myself one of the cheaper certificates for a few dollars a year from SSL Shopper. And I gathered all this information over a week of looking things up in my free time. I was quite sure that I had covered all viable options but I couldn’t help wondering whether it’s possible to get an SSL certificate for my personal domain for free. One final DuckDuckGo search led me to a StackOverflow post that answered this question in affirmative! Let’s Encrypt The StackOverflow post pointed me to the Let’s Encrypt initiative which essentially offers means for one to generate SSL certificate oneself via a fully automated verification process. Not just that, it offers you with a host of ways in which this can be done depending on your level of comfort with using the command-line, cPanel or any other means through which verification can be carried out. I was skeptical that something like this is too good to be true, but it isn’t. Also, the project is sponsored by several well-known organisations such as the Linux Foundation, Mozilla, EFF and CISCO. And the certificate is accepted by all mainstream browsers. As a coincidence, I later found out that The Site Wizard, which I had referred to several times in the past while choosing a hosting provider, website templates, etc. is also secured by a certificate from Let’s Encrypt! Now this was exactly what I wanted, i.e. to secure my personal domain so that I can transfer data between my location and my ownCloud server. It does not matter to me (at least for now) how much extra assurance a seal from a known CA such as DigiCert or Symantec gives a visitor to my website. Plus, it’s absolutely free. In my case, I had the certificate generated within minutes through ZeroSSL with an automated ACME verification process that involved me creating two files with specific content on my domain that were verified by this website. There are many alternatives to ZeroSSL, any of which can be used as per one’s convenience. One thing to note is that the certificate issued by ZeroSSL is valid only for three months, but I don’t mind repeating the very simple process again when my current certificate expires. Last Words So to conclude, securing one’s website with a TLS/SSL certificate is not as hard or expensive as it may seem at first glance thanks to Let’s Encrypt. I’m very impressed by this initiative, and found it to be a perfect alternative for my needs given all other options known to me. The Let’s Encrypt team is currently seeking funding for their operations and I’m about to donate to it as a token of my appreciation. So if you are in a similar situation as I was before my research that led me to Let’s Encrypt, I hope you benefit from reading this post!

A Visit to MusicMuni Labs

Mon, 12 Dec 2016 00:00:00 +0000

While Nina presented on Music Therapy for Dementia at ARDSICON 2016 in Bangalore, I visited MusicMuni Labs — a startup founded by Gopala Koduri and Sankalp Gulati, mentored by Prof. Xavier Serra and researchers from the Music Technology Group at Universitat Pompeu Fabra.

They’re building on research from the CompMusic project — focused on Hindustani and Carnatic classical music — with two products currently in beta:

Riyaz — An Android app for beginner to intermediate students that uses music technology to evaluate singing against reference lessons and give detailed feedback.

Jukedeck at the Science Museum Lates

Thu, 01 Sep 2016 00:00:00 +0000

Jukedeck had a stall at the Science Museum’s Lates — a monthly adults-only after-hours event in London — and I joined colleagues Patrick, Lydia, Eliza, Matt, Katerina, and Gabriele to demonstrate the music generation technology.

The museum was packed. It felt like a poster presentation at a conference, but with a non-technical audience — and a constant stream of curious visitors with barely a moment for a break. For someone who works on the research and code behind the music generation, seeing how the public engaged with it was genuinely rewarding.

Music and Connectionism

Mon, 01 Aug 2016 00:00:00 +0000

The many contributions made during the past three decades to computer-assisted analysis and generation of music with the aid of Connectionist architectures can be seen to have occured in two waves, in parallel with developments in Connectionist research itself. During the first wave, the founding principles of Connectionism were introduced (Rumelhart et al., 1986) through the idea of Parallel Distributed Processing according to which mental phenomena occur as a result of simultaneous interactions between simple elementary processing units, as opposed to the then prevailing notion of Sequential Symbolic Processing which explained the same phenomena in terms of sequential interactions between complex goal-specific units. Its significance is largely theoretical, with a few experimental and empirical results to support the feasibility of the theory. Following several years of reduced interest, the second wave further strengthened the claims made by its precursor through a series of successful high-impact real-world applications. This was owing to both the proposal of newer theories, and the availability of greater computational power and vast amounts of data that enabled the demonstration of the efficacy of these theories nearly two decades on (Bengio, 2009; LeCun et al.,2012). The innovations that came about as a result of these two phases trickled down to several application domains (Krizhevsky et al., 2012; Hinton et al., 2012;Collobert et al., 2011) of which music is one (Todd and Loy, 1991; Griffith and Todd,1999; Humphrey et al., 2012). This section reviews notable contributions among the many that demonstrated the application of connectionism to symbolic music modelling during these two waves in order to present a historical perspective together with an overview of the techniques employed.

New Guitar Video – Them Bones by Alice in Chains

Sat, 26 Sep 2015 00:00:00 +0000

Them Bones opens Dirt (1992) with one of the most distinctive riffs in the grunge canon — a lurching 7/8 groove that gives the whole song an unsettling, forward-leaning feel. This was recorded not long after my Man in the Box cover, continuing a run through the Alice in Chains back catalogue.

New Guitar Video – Vicarious by Tool

Tue, 08 Sep 2015 00:00:00 +0000

Vicarious opens 10,000 Days (2006) and was one of the first Tool songs I worked through properly. The rhythmic complexity is characteristic Tool — a recurring 5-against-4 feel woven through an otherwise driving riff. Not their most technically demanding song to play, but one of the most satisfying.

Oral Presentation at IJCNN 2015

Mon, 20 Jul 2015 00:00:00 +0000

My paper was accepted for oral presentation at the 28th International Joint Conference on Neural Networks in Killarney, Ireland.

“Discriminative Learning and Inference in the Recurrent Temporal RBM for Melody Modelling” — The paper proposes the Recurrent Temporal Discriminative RBM (RTDRBM), which combines discriminative learning with recurrent structure to predict probability distributions for the next note in a melody. Evaluated on 8 folk and chorale melody datasets, it outperforms n-grams and other connectionist models with statistically significant improvements.

Oral Presentation at ISMIR 2014

Tue, 04 Nov 2014 00:00:00 +0000

I attended ISMIR 2014 in Taipei with two accepted papers:

“Multiple Viewpoint Melodic Prediction with Fixed-Context Neural Networks” — Continuing earlier work on neural network–based note prediction using the multiple viewpoints representation, an event-based representation of symbolic music data.

“An RNN-based Music Language Model for Improving Automatic Music Transcription” — Co-authored with Siddharth Sigtia and Emmanouil Benetos.

Both presented as posters. Prof. Ye Wang’s keynote on music’s therapeutic applications in rehabilitation was a highlight — it made me think about how melody modelling research could eventually serve therapeutic purposes.

Poster Presentation at the Machine Learning Summer School

Fri, 09 May 2014 00:00:00 +0000

I was selected to attend the Machine Learning Summer School in Reykjavik (April 25–May 4, 2014) and received a travel grant. I presented a poster on musical pitch prediction with neural networks.

Particularly valuable tutorials: Machine Learning and HCI (Roderick Murray-Smith), Introduction to ML (Neil Lawrence), Deep Learning (Yoshua Bengio). The reinforcement learning content was especially interesting — I could see potential applications in my music modelling work.

Outside the summer school: the Golden Circle Tour, a hike up Mount Esjan, and a last-minute trip to the Blue Lagoon before flying home. Reykjavik is one of the most unique places I’ve visited.

Oral Presentation at ISMIR 2013

Sun, 17 Nov 2013 00:00:00 +0000

I presented “A Distributed Model for Multiple Viewpoint Melodic Prediction” at ISMIR 2013 in Curitiba, Brazil. The paper uses a Restricted Boltzmann Machine to model melodic sequences — demonstrating that the model can make use of information in longer contexts more effectively than n-gram models. It won the Best Student Paper Award.

I also organised a late-breaking session on MIR applications in educational settings.

After the conference I spent a week in Rio de Janeiro — Cristo Redentor, Sugarloaf Mountain, beaches near Copacabana. My supervisor and a colleague came along for sightseeing. A fabulous experience.

Oral Presentation at the 6th International Workshop on Machine Learning and Music

Tue, 24 Sep 2013 00:00:00 +0000

Two papers accepted at MML 2013, co-located with the European Conference on Machine Learning.

“A Neural Probabilistic Model for Predicting Melodic Sequences” — Uses RBMs for folk melody classification, outperforming n-gram models, with linear scaling with context length.

“An Efficient Shift-Invariant Model for Polyphonic Music Transcription” — Co-authored with Emmanouil Benetos.

I couldn’t attend, so Emmanouil presented both. Slides and PDFs available on request.

Downloads

MML 2013 Paper MML 2013 Presentation MML 2013 Paper

Poster Presentation at the 3rd Annual Researchers' Symposium

Thu, 27 Jun 2013 00:00:00 +0000

I presented “A Neural Network for Predicting Musical Pitch” at City University London’s 3rd Annual Researchers’ Symposium — an event where doctoral students present their research to non-specialist audiences.

The work uses RBMs to predict sequences of musical pitch from monophonic MIDI melodies, outperforming n-gram models while requiring fewer parameters. The poster won the Best Poster Presentation Award.

Poster created in Beamer/LaTeX — PDF available on request.

City University 3rd Annual Researchers’ Symposium Poster

Oral Presentation at the 5th BCS Doctoral Consortium

Fri, 17 May 2013 00:00:00 +0000

My abstract was accepted for the BCS 5th Doctoral Consortium and I presented on May 16, 2013.

“A Neural Probabilistic Model for Music Prediction” — Proposes using a Restricted Boltzmann Machine to model musical pitch sequences in monophonic MIDI melodies, addressing the context limitations of Markov models. The model can make use of information in longer sequences more effectively than recently evaluated Markov models, with potential applications in music education, composition, transcription, and classification.

Code

Mon, 01 Jan 0001 00:00:00 +0000

Research implementations and experiments. Most of these accompany published papers.

Discriminative Restricted Boltzmann Machine An implementation of the Discriminative RBM (DRBM) in Python using Theano.

Generalised Discriminative Restricted Boltzmann Machine An implementation of the generalised DRBM in Python using Theano, as proposed in Generalising the Discriminative RBM (ICANN 2017).

Neural Probabilistic Language Model A basic implementation of the Neural Probabilistic Language Model in Python using Scikit-learn.

TensorFlow Datasets API Examples Example scripts for using the TensorFlow Datasets API to train Recurrent Neural Networks.

Experience

Mon, 01 Jan 0001 00:00:00 +0000

Senior Machine Learning Developer 2020 – present

Unity Technologies · Copenhagen, Denmark

AI tools for game developers across multiple product areas. Built gameplay simulation agents for modelling player behaviour in game economies, and researched diffusion models for visual asset generation. Worked on ads pipeline monitoring before joining the Unity Muse team — an LLM-driven AI assistant integrated into the Unity Editor — in March 2023.
Senior Researcher 2018–2020

Moodagent · Copenhagen, Denmark

Music technology research focused on personalisation. Developed music recommendation systems using collaborative filtering based on user interaction patterns with the Moodagent streaming app.

Music

Mon, 01 Jan 0001 00:00:00 +0000

Music has been a constant throughout my life, both as a profession and a passion.

Guitar

I play electric and acoustic guitar, mostly rock and heavy metal. I post covers to my YouTube channel, these_riffing_hands — subscribe if you’d like to follow along.

I hold Grade 6 certification in Electric Guitar (Rock School) and have informally worked through the higher grades.

Compositions

I share original compositions and collaborations on SoundCloud — link coming soon.

Other

Mon, 01 Jan 0001 00:00:00 +0000

Reviewing

ISMIR — 2014, 2015, 2018, 2019
IJCAI — Sub-reviewer, 2015
Frontiers in Digital Humanities — Review Editor, Digital Musicology

Mentoring

Mentor in the Widening Inclusion in Music Information Retrieval (WiMIR) programme — 2018, 2019, 2020, 2025, 2026.

Supervision

Sezal Jain — Siemens Corporate Technology
Vaibhav Kumar — City, University of London

Events

Program Chair — City Informatics PhD Symposium, 2014
Participant — Schloss Dagstuhl Seminar on Neural Symbolic Learning
Panelist — Computer Simulation of Musical Creativity Conference: “Applying Music Patterns in Generation”

Publications

Mon, 01 Jan 0001 00:00:00 +0000

2018

StructureNet: Inducing Structure in Generated Melodies Medeot G., Cherla S., Kosta K., McVicar M., Abdallah S., Selvi M., Newton-Rex E., Webster K. International Society for Music Information Retrieval Conference (ISMIR) Jukedeck Ltd.

Linear-Time Sequence Classification using Restricted Boltzmann Machines Tran S., Cherla S., Garcez A., Weyde T. arXiv Preprint — arxiv.org/abs/1710.02245 City, University of London

2017

Generalising the Discriminative Restricted Boltzmann Machine International Conference on Artificial Neural Networks (ICANN) City, University of London