Skip to content
Neural Aspect

Research

Notes on our prediction work: how numbers change over time, and what a table of past results says will happen next. Older research notes stay here too.

Let’s build a Salesforce Deal Predictor (Part 1)

research

Let’s build a Salesforce Deal Predictor (Part 1)

Salesforce has been the darling of the CRM world for as long as I can remember. Businesses of all shapes and sizes use it to manage their sales processes. For some time their built-in opportunity scoring product has been predicting outcomes of deals, but anecdotally the clients I have spoken to have been underwhelmed by it’s performance. So we’re going to put that right in this article and explore an alternative way of generating predictions on opportunities … the point is not so much to give you an assignment to create your own solution, but rather to give you a sense of how the solution is a combination of data and AI and what’s important in any custom model AI project that you embark on.

How to optimise your marketing spend with Meridian

research

How to optimise your marketing spend with Meridian

Optimising Marketing Spend with Google's MeridianGoogle's new Marketing Mix Model, Meridian, offers a fresh approach to understanding and optimising marketing investments. In my latest article, I explore how Meridian leverages Bayesian inference to provide actionable insights into channel performance, ROI, and budget allocation.​While it requires some technical setup, the potential benefits for strategic decision-making are substantial.​If you're interested in data-driven strategies to enhance your marketing effectiveness, this piece provides a practical overview.​

Why Your Sales Forecasting Fails: The Critical Role of Time in CRM Data and Machine Learning

research

Why Your Sales Forecasting Fails: The Critical Role of Time in CRM Data and Machine Learning

Most sales teams are sitting on a goldmine of data—but their machine learning models are digging in the wrong place.In this article, we reveal why sales forecasting often falls short, the hidden flaw in CRM data that’s undermining your insights, and how to fix it.

Deep Dive of Google’s TSMixer

research

Deep Dive of Google’s TSMixer

A deep dive into Google’s TSMixer paper

9 Pitfalls to avoid predicting Sales Opportunities

research

9 Pitfalls to avoid predicting Sales Opportunities

In this post, we’ll explore some of the common challenges that can derail your efforts to predict sales opportunities accurately. Drawing from experience, I’ll highlight potential solutions to avoid these pitfalls and build a more reliable system.

Our Top 7 Forecasting Models We Benchmarked For Monash

research

Our Top 7 Forecasting Models We Benchmarked For Monash

Top 7 Forecasting Models We Benchmarked For Monash

Revin: What really grinds my gears

research

Revin: What really grinds my gears

A review of Revin - Reversible Instance Normalization used in time series forecasting

Multivariate Time Series models: Do we really need them?

research

Multivariate Time Series models: Do we really need them?

A comparison of local, global, univariate and multivariate configurations using the DLinear and NLinear models

LTSF-Linear: Embarrassingly simple time series forecasting models

research

LTSF-Linear: Embarrassingly simple time series forecasting models

A review of the 2022 Paper Are Transformers Effective for Time Series Forecasting that introduced DLinear and NLinear models

Pytorch vs MXNet: Which is faster?

research

Pytorch vs MXNet: Which is faster?

An analysis of the computational efficiency of Pytorch and MXNet

The F1 Score: Time Series Model Championships

research

The F1 Score: Time Series Model Championships

A ranking system of time series models based on the Monash dataset benchmarks using the mase metric and the formula 1 scoring system.

1Cycle scheduling: State of the art timeseries forecasting

research

1Cycle scheduling: State of the art timeseries forecasting

How to get state of the art timeseries forecasting results using machine learning with my variant of DeepAR and 1Cycle Scheduling

Super-convergence: Supercharge your Neural Networks

research

Super-convergence: Supercharge your Neural Networks

A look at 1Cycle scheduling, one of my favourite techniques at improving model performance and practical guidance on how to use it

A fistful of MASE: Deconstructing DeepAR

research

A fistful of MASE: Deconstructing DeepAR

A deep dive into the GluonTS DeepAR neural network model architecture for time series forecasting and an ablation study of the covariates.

Predicting covariates: Is it a good idea?

research

Predicting covariates: Is it a good idea?

A study which evaluates the effectiveness of predicting covariates in LSTM Neural Networks for Time Series Forecasting

Teacher Forcing: A look at what it is and the alternatives

research

Teacher Forcing: A look at what it is and the alternatives

Review of teacher forcing, free running, scheduled-sampling, professor forcing and attention forcing for training auto-regresssive neural networks

Human-level control through deep reinforcement learning

research

Human-level control through deep reinforcement learning

Recreating the experiments from the classic 2015 Deepmind Paper by Mnih et al.: Human-level control through deep reinforcement learning

Revisiting Playing Atari with Deep Reinforcement Learning

research

Revisiting Playing Atari with Deep Reinforcement Learning

Recreating the experiments from the classic DQN Deepmind paper by Mnih et al.: Playing Atari with Deep Reinforcement Learning