### Where I give some details about the project and my background

And also to put it in a simple way, I cite the quotes from James Bond’s movie Skyfall between Q and James Bond in the London underground. Being asked to get on the (moving) train, Bond did of course get on the train. At Q’s later question “Where are you?”, the 007 agent replies: “take a wild guess, Q”. Enjoy this amazing 1-minute scene here.

This platform wants to develop intelligent, smart, adaptive agents that could maximize their reward in a complex environment. The expression take a wild guess comes from the idea that humans could feel most of the time what is the right thing to do in a given scenario and condition and keep taking the proper decision over the time.

#### Mattia Venditti

I am passionate about AI and I have been developing some applications that could benefit from this new technique.

One of the first areas I started investigating the AI potential was my PhD research field. My work mainly focused on developing a simulation environment for Plug-in Hybrid Electric Vehicles and Hybrid Powertrain Optimal Control. After completing it in 2015, my goal as post-doc researcher was to develop machine learning algorithms for advance vehicle control, exploiting Vehicle-to-Infrastructure data. My focus was on the study of energy consumption, $CO_2$ emissions and costs of Plug-in Hybrid Electric vehicles and Full Electric Vehicles. During those years, I also followed several projects about urban mobility with international partners from industry and academia.

One different field I have been delving into is Natural Language Understanding. In fact, I have been developing a semantic engine for the last four years. The engine interprets a natural language query and translates it into an equivalent formal query (such as SPARQL) that retrieves the answer from a Knowledge Base (such as FreeBase). The framework is built in Python.

### My interests

##### Artificial intelligence.

I am currently developing machine learning algorithms in adversarial/cooperative environments of two or more agents.

##### Autonomous driving

I am currently developing deep learning algorithms that maximize self-driving car safety in a simulated environment, such as video games. One of the major concerns about autonomous driving vehicles should be to improve energy efficiency, in addition to safety. I intend to develop a simulation framework where dedicated reinforcement learning algorithms aim at identifying the vehicle velocity trajectories, given the traffic conditions, that minimize energy consumption of different power-train solutions.

##### Semantic Engine

I have been developing a semantic engine (as a side project) for the last four years. The engine goal is to interpret a natural language query and to translate it into an equivalent formal query (such as SPARQL) that retrieves the answer from a Knowledge Base (such as FreeBase). I have been building the framework in Python, running on Ubuntu. The entire application back-end is written in Flask, the training procedure has been initially carried out using Genetic Algorithms implemented in NumPy. I then started investigating Deep Learning and TensorFlow as development framework to implement a Neural Programmer Interpreter.

##### Language Modelling

I am applying new parsing techniques with the support of encoding networks to understand/generate text. I am developing an extension of the Skip-gram model to learn multi word embeddings per word, with respect to the relation-type set of a given Knowledge Base. I intend to apply state-of-art techniques to Knowledge Base Completion. The knowledge base capacity is a key factor for a large applicability of a semantic engine.

### Publications

Here I list my main publications during my PhD and post-doc research.

2016

Elsevier, Applied Energy

Tools for cost-optimized design and performance of HEVs. Dual-mode parallel hybrid architecture. Optimal control strategy based on fuel consumption, NOx and battery aging. Impact of cost definitions on HEV design. Impact of fuel & battery price and battery life on cost-optimized design of HEVs

2016

Elsevier, Energy Conversion and Management

Non-plug-in dual-mode parallel hybrid architecture. Cross-validation machine-learning for robust equivalent consumption-based controllers. Optimal control strategy based on fuel consumption, NOx and battery aging. Impact of different equivalent consumption definitions on HEV performance. Correlation between vehicle braking energy and SOC variation in the traction stages.

2014

Elsevier, Applied Energy

Layout design, energetic and cost analysis of complex parallel hybrid vehicles. Development of global and real-time optimizers for control strategy identification. Rule-based control strategies to minimize fuel consumption and NOx. Energy share across each working mode for battery and thermal engine.

2016 Sep. 14-16 71st Congress ATI 2016, Torino, Italy

Elsevier, Energy Procedia

2016 Apr. 12-14 SAE 2016 World Congress & Exhibition, Detroit, Michigan, US

SAE, International Journal of Alternative Powertrains

### Talks

Here I list my main talks at the Machine-Learning meetup in Torino, Italy.

May 10, 2018 Torino, Italy

This meetup has been about the implementation of a machine learning algorithm in Python for hybrid electric vehicle applications. The algorithm is expected to learn how to manage the motors of the hybrid powertrain to reduce pollutant and CO2 emissions, in many different driving scenarios.