NVIDIA DLI - Capital Markets

Nvidia DLI logo

The NVIDIA Deep Learning Institute (DLI) offers hands-on training for developers, data scientists, and researchers looking to solve the world’s most challenging problems with deep learning.

Through self-paced online labs and instructor-led workshops, DLI provides training on the latest techniques for designing, training, and deploying neural networks across a variety of application domains. Students will explore widely used open-source frameworks as well as NVIDIA’s latest GPU-accelerated deep learning platforms.

workshops logo

During the Workshop Day (December 5) at NMAI we will be offering a full day with the leading NVIDIA trainers for the financial services sector.  The day will include a general introduction session, followed by Labs #1 and #2, and then finally at the end of the day we will provide a taster of the yet to be released Lab #3.

The day is inclusive of a light breakfast, refreshments throughout the day, lunch and an invitation to the main conference Icebreaker Reception, taking place at the same venue between 6-8pm.  This reception will be open all all workshop participants, plus the speakers and delegates from the main conference.

Entry into the main conference is not included.

Purchase tickets

DLI Finance Lab #1

Prediction of Time Series Financial Data using LSTM for Algorithmic Trading (Level intermediate)

Lab abstract:

This Lab demonstrates how to structure and train LSTM deep neural networks to predict time series behavior using technical and fundamental inputs. This lab is based on the dataset in the Kaggle contest called the "Two Sigma Financial Modeling Challenge", and contains anonymized features pertaining to a time-varying value for financial instruments. We use the LSTM network to train a predictor of the target variable optimized to find the highest possible correlation with the labelled targets.  The lab uses the TensorFlow deep learning framework and Python data science tools like Pandas to perform data cleaning, RNN network construction, training, and evaluation.

After taking this lab you will be able to:

  • Structure and train an LSTM network in TensorFlow to accept vector inputs and predict a target
  • Prepare time series data and test network performance using training and test datasets
  • Understand the steps in creating an end-to-end RNN time series prediction algorithm in TensorFlow that could be benchmarked against traditional machine learning techniques



  • Working knowledge of basic scientific python
  • Basic level knowledge of TensorFlow

DLI Finance Lab #2 

Deep Autoencoder based Statistical Arbitrage Strategy (Level intermediate)

Lab abstract:

Linear techniques such as PCA are the workhorse of creating ‘eigenportfolios’ that can be used for statistical arbitrage strategies. This lab demonstrates a new method of using a deep autoencoder as an alternate method that can be used to reconstruct security return data, and thus serve as a basis for performing statistical arbitrage using non-linear deep learning.  Anomalous deviations (spreads) in reconstruction error are used as (mean reverting) signals for creating long/short positions.  P&L of the strategy is calculated and thus can be benchmarked against traditional linear techniques. This course uses TensorFlow and Python for the code examples covered during the lab session.

After taking this lab you will be able to:

  • Structure and train a deep autoencoder in TensorFlow and Python.
  • Use the autoencoder as an anomaly detector to create an arbitrage strategy and perform hyperparameter optimization over the autoencoder.
  • Calculate P&L of the strategy



  • Working knowledge of basic scientific python
  • Basic level knowledge of TensorFlow
  • Knowledge of PCA techniques for statistical arbitrage

DLI Finance Lab #3

Deep Reinforcement Learning for Optimizing Large Order Trade Execution (Level advanced)

Lab abstract:

This Lab demonstrates how to use Deep Reinforcement Learning (D-RL) for optimizing the execution of large security orders. A DNN-based policy is used to implement GA3C actor-critic model approach.  A simple market simulator is used to capture the effects of temporary and permanent impact functions, based on some generally accepted features such as current and lagged values of spread, best bid/ask, volume, volatility, log-return, bid-ask imbalance, etc. Simulated implementation shortfalls can then be calculated for various trading trajectories, thus providing a penalty/reward function. The D-RL GA3C technique is used to find a policy that can generate optimal trading trajectories that minimize implementation shortfall, and can be benchmarked against other well-known methods.  This approach should provide a starting framework for incorporating higher dimensional order book data into the methodology in a way that would allow state-of-the-art optimization networks to be developed.

After taking this lab you will be able to:

  • Understand deep reinforcement learning via a GA3C actor-critic technique implemented as a DNN policy, based on implementation shortfall as the penalty function
  • Understand how D-RL integrates with a simple market simulator as the system environment, providing a platform for experimentation and extension



  • Working knowledge of scientific python
  • Intermediate level knowledge of TensorFlow
  • Some prior understanding of policy gradient deep reinforcement learning techniques
  • Working knowledge of trade execution trajectories, impact functions, and financial time-series models such as GARCH and ARIMA