A look back at Synapse, Italy’s first AI symposium

Last month, leading voices in the global AI research community gathered in Milan, Italy, to discuss advances in the field — and to weigh in on some of the most pressing questions facing humanity in the age of machine learning. Synapse, the symposium on artificial intelligence hosted by Bending Spoons, was the first event of its kind to take place in Italy and drew over 150 of the brightest computer science students from around the world.

At the cutting edge of AI research

The invited speakers represented the vanguard of AI research — hailing from some of the world’s most prestigious entities and institutions such as Google, Meta, MIT, and the École Polytechnique Fédérale de Lausanne. Diving deep into the machine learning innovations shaking up the field, they covered everything from ethically aware AI to AI that can paint, write source code, and train itself. In other words, AI innovations that are no longer the stuff of imagination, but visible on a fast-approaching horizon — if not already the reality.

The event likewise shed light on some of the most compelling questions driving the direction of AI research today. Questions such as “How does AI learn to perceive the real world?” “What’s preventing the roll-out of self-driving cars?” and perhaps most crucial of all, “What needs to be done to ensure AI creates a more equal and inclusive society?” Questions that, once answered, will undoubtedly lead to new questions, new goals, and new benchmarks for what is possible in the age of artificial intelligence.

A who’s who of AI innovators

Speaking on the topics that fascinate them most, the experts shared their unique perspectives and findings from their latest research.

Max Tegmark
Opportunities and challenges in the age of AI

Max Tegmark — a professor and researcher in AI and physics at MIT, as well as the president of the Future of Life Institute and author of two bestsellers and over 250 publications — was the first thought leader to speak at Synapse. Setting the tone for the event, Tegmark gave a forward-looking keynote on the future of AI — the opportunities it promises, the challenges it will bring, and its place in society.

Addressing the audience of bright computer science students on the brink of exciting — and in all likelihood, impactful — careers, Tegmark spoke of the responsibility that current and future generations of AI innovators must bear: “You are key in this. I would encourage you all to envision how we can reinvent — reimagine — how AI can be developed so that it remains aligned with what is actually good for all people.”

Vittorio Ferrari
Transformers in multi-object 3D reconstruction

A senior staff research scientist at Google leading a research group on visual learning, Vittorio Ferrari shared insights into the latest developments in 3D deep learning — including the groundbreaking work he and his team are doing to improve both the process and results of 3D reconstruction from a single RGB image. Ferrari also introduced the audience to a novel approach to multi-object 3D reconstruction: RayTran — a two-stream neural network that gradually consolidates 2D features on a 3D volume, and vice versa, with repeated blocks of ray-traced transformers.

Luigi Di Stefano
Multi-task learning and self-supervision in stereo vision

A full professor and the founder and leader of the Computer Vision Lab at the University of Bologna, Luigi Di Stefano joined the event as a pioneer in computer stereo vision. He discussed the prohibitive limitations of collecting large-scale annotated datasets for supervised learning models in dense stereo vision tasks — and the work he and his research team are doing to address them. Among the solutions, Di Stefano described a multi-task learning approach and self-supervised training in which no human annotation is required.

Alexandre Alahi
Socially aware AI in self-driving cars

Assistant Professor at the EPFL and winner of the CVPR Open Source Award and ICDSC Challenge Prize, Alexandre Alahi presented on visual intelligence for transportation. According to Alahi, a specialist in the theoretical challenges and practical applications of socially aware AI, self-driving cars must be able to forecast future human behavior — pedestrian movement and intent. To achieve this level of social awareness in AI, Alahi proposed a paradigm shift in how it’s trained — utilizing both knowledge-driven models (such as Kalman filters and linear trajectory avoidance models) for more robustness and interpretability and data-driven models (with data on human social interactions) for greater accuracy.

Georgia Gkioxari
Leveraging 2D supervision to get to 3D vision

Meta superstar research scientist and recipient of the 2021 PAMI Young Researcher Award, Georgia Gkioxari’s Synapse talk underscored her place at the forefront of machine learning research for computer vision. Her latest work on AI visual perception — and her presentation topic at Synapse — addressed the issue of the unscalability of using 3D supervision to train models to predict 3D shape and layout. Emphasizing the importance of having vast quantities of data and various sensors in computer vision, Gkioxari described a new method that learns to predict shape and layout for 3D objects relying only on multi-view images with 2D supervision.

Michele Catasta
Source code and scaling large language models

Head of applied research at X, the moonshot factory (formerly Google[x]), Michele Catasta spoke about his current work, which challenges the status quo of source code creation and investigates how AI can revolutionize the way software is developed. Rather than replacing engineers, argued Catasta, AI would make them even more productive. A specialist in large language models, Catasta further discussed the vast untapped potential of LLMs for natural language tasks — even source code generation. According to him, using transformers to scale LLMs would have an unprecedented impact on virtually countless language-related applications.

Federico Perazzi
Generative models for the creation of visual content

Head of AI at Bending Spoons and a research scientist specializing in AI-powered computer vision tools for professional image and video editing, Federico Perazzi illustrated how machine learning has entered the realm of artistic expression. Presenting on the topic of AI in the creation of visual content, Perazzi discussed the evolution of generative models and revealed how these models work. He further demonstrated the underlying technologies that enable the transformation of images from low resolution to high resolution — and natural-sounding text descriptions to never-before-generated images.

Poster sessions — from algorithmic fairness to automatic road safety assessment

In addition to the talks, Synapse featured the recent research work of several PhD candidates and Bending Spoons AI researchers in the form of poster sessions. Attendees could view the posters and speak with the researchers during breaks throughout the event.

Here’s the complete list of the researchers and their poster topics:

• Alberto Parravicini and Davide Carletti, “Faster-than-Light On-Device Shot-Cut Detection”
• Alessandro Berlati and Massimo Zambelli, “Feeding Research With Good Data. A Lot of It.”
• Alessandro Serena, “A Lightweight Typing Assistant for Fonts”
• Davide Paglieri, “SlowMo.ai: Video Frame Interpolation”
• Francesco Giuliari, “Spatial Commonsense Graph for Object Localisation in Partial Scenes”
• Jenny Yang, “Achieving Algorithmic Fairness through Adversarial Training”
• Juan Manuel Molina Maza, “Convolutional Neural Network Scheme for Pediatric Brain MRI Super-Resolution”
• Karolina Pakenaite, “Communicating Photographic Content through Tactile Images”
• Lorenzo Collodi, Andrea Pesare, and Jelena Rodic, “Blind Image Restoration”
• Luca Actis Grosso, Noe Guisset, and Sergiu Neamtu, “Orion API: Build, Train, and Scale Models. Fast.”
• Marco Aspromonte, “Neural Radiance Field (NeRF) for View Synthesis of Arbitrary Objects”
• Marin Kačan, “End-to-End Automatic Road Safety Assessment”
• Monica Romero, “Improving Aphasic Speech Recognition through Novel Semi-Supervised Learning Methods on AphasiaBank for English and Spanish”
• Pietro Fanti and Andrea Ferretti, “Color Enhancement”
• Riccardo Albertazzi and Lorenzo Migliorino, “Model Serving Made Easy”
• Safa Hamreras, “Content-Based Image Retrieval by Ensembles of Deep Learning Object Classifiers”

The significance of an AI event in Italy

In his welcome speech, Bending Spoons CEO Luca Ferrari spoke about the Italian tech company’s view on AI’s tremendous role in shaping the world — and its bearing on society. “We envision a brighter future of AI, in which the knowledge and control of AI are a global endeavor and AI is used to benefit humanity as a whole.”

But such a future is not guaranteed. Especially when considering that, as Ferrari pointed out, the lion’s share of AI knowledge and control is concentrated in very few places in the world — making an AI event held in Italy especially significant. Because if left unchecked, Ferrari explained, the concentration of power will only grow — along with the risk that AI will lead to an increasingly unequal society with a miserable, excluded majority. The hope, said Ferrari, is that events such as Synapse AI Symposium will “help ignite a movement in favor of introducing a global system of checks and balances to maximize the probability that AI makes our world more equal and inclusive than ever before.”

In collaboration with the African Institute for Mathematical Sciences (AIMS)

With the conviction that AI research must be a global endeavor, Bending Spoons collaborated on Synapse with AIMS — Africa’s first and most extensive network of academic centers focused on empowering Africa’s youth to “shape the continent’s future through STEM education, public engagement, and research.” Bending Spoons brought several AIMS students to Italy to participate in the event, engage the speakers in conversion, and network with their STEM peers from Europe and beyond.

The overall impact of AI looks promising–if we tread carefully

There was a palpable sense of excitement at Synapse — about the progress of AI and the new possibilities emerging as machine learning becomes ever more advanced. There was also a sense of cautious optimism about the future. The speakers were careful to point out potential risks and misuses of AI, and to emphasize the importance of mitigating these risks to ensure a positive outcome for society at large.

Because if there was one universal message to emerge from Synapse, it was this: AI will fundamentally and irrevocably change the world, bringing enormous consequences to all reaches of humanity. Whether that change will be for the good of all is being determined in large part by the choices, considerations, and recalibrations we make at every step of AI research and development.

So we cannot afford to underestimate the weight of our responsibility as members of the greater AI community — and as members of a global society.

A final word on the success of the first AI event in Italy

From the world-class status of the guest speakers to the attending students’ brightness, curiosity, and drive, the success of Synapse can be measured by the caliber of its participants — and their enthusiastic response to the event. It’s a success that proves there is not only a place for AI knowledge-sharing in a smaller country but also a demand for it. And as discussed, there is also a moral obligation for it. Which is why Bending Spoons will continue to facilitate inclusiveness, representation, and knowledge-sharing within the global AI research community in future events and initiatives.