Nicolò Savioli, Ph.D.

AI Italian Researcher 🇮🇹 | PhD @ King’s College London | AI feels stagnant—papers keep recycling the same ideas, now tweeting out of sheer boredom

[Computer Vision and Pattern Recognition Papers @CSVisionPapers]

Venice-H1: Failure-Aware Query Re-Ranking with Multi-Scale Grid Signatures for Referring Image Segmentation

Nicolò Savioli
https://t.co/5039hBlmCw [𝚌𝚜.𝙲𝚅]
💬Code: https://t.co/DT2P0QFqvi

[Nicolò Savioli, Ph.D. @NSavioli]

@Michael_J_Black Diffusion is not a scalable model.

[Nicolò Savioli, Ph.D. @NSavioli]

@SchmidhuberAI @predict_addict Feed-forward neural nets are the core limitation of today’s AI; backprop is just their byproduct. We need a new, biologically inspired paradigm with truly adaptive, self-organizing networks.

[Jürgen Schmidhuber @SchmidhuberAI]

Who invented convolutional neural networks (CNNs)?

1969: Fukushima had CNN-relevant ReLUs [2].

1979: Fukushima had the basic CNN architecture with convolution layers and downsampling layers [1]. Compute was 100 x more costly than in 1989, and a billion x more costly than today.

1987: Waibel applied Linnainmaa's 1970 backpropagation [3] to weight-sharing TDNNs with 1-dimensional convolutions [4].

1988: Wei Zhang et al. applied "modern" backprop-trained 2-dimensional CNNs to character recognition [5].

All of the above was published in Japan 1979-1988.

1989: LeCun et al. applied CNNs again to character recognition (zip codes) [6,10].

1990-93: Fukushima’s downsampling based on spatial averaging [1] was replaced by max-pooling for 1-D TDNNs (Yamaguchi et al.) [7] and 2-D CNNs (Weng et al.) [8].

2011: Much later, my team with Dan Ciresan made max-pooling CNNs really fast on NVIDIA GPUs. In 2011, DanNet achieved the first superhuman pattern recognition result [9]. For a while, it enjoyed a monopoly: from May 2011 to Sept 2012, DanNet won every image recognition challenge it entered, 4 of them in a row. Admittedly, however, this was mostly about engineering & scaling up the basic insights from the previous millennium, profiting from much faster hardware.

Some "AI experts" claim that "making CNNs work" (e.g., [5,6,9]) was as important as inventing them. But "making them work" largely depended on whether your lab was rich enough to buy the latest computers required to scale up the original work. It's the same as today. Basic research vs engineering/development - the R vs the D in R&D.

REFERENCES

[1] K. Fukushima (1979). Neural network model for a mechanism of pattern recognition unaffected by shift in position — Neocognitron. Trans. IECE, vol. J62-A, no. 10, pp. 658-665, 1979.

[2] K. Fukushima (1969). Visual feature extraction by a multilayered network of analog threshold elements. IEEE Transactions on Systems Science and Cybernetics. 5 (4): 322-333. This work introduced rectified linear units (ReLUs), now used in many CNNs.

[3] S. Linnainmaa (1970). Master's Thesis, Univ. Helsinki, 1970. The first publication on "modern" backpropagation, also known as the reverse mode of automatic differentiation. (See Schmidhuber's well-known backpropagation overview: "Who Invented Backpropagation?")

[4] A. Waibel. Phoneme Recognition Using Time-Delay Neural Networks. Meeting of IEICE, Tokyo, Japan, 1987. Backpropagation for a weight-sharing TDNN with 1-dimensional convolutions.

[5] W. Zhang, J. Tanida, K. Itoh, Y. Ichioka. Shift-invariant pattern recognition neural network and its optical architecture. Proc. Annual Conference of the Japan Society of Applied Physics, 1988. First backpropagation-trained 2-dimensional CNN, with applications to English character recognition.

[6] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, L. D. Jackel: Backpropagation Applied to Handwritten Zip Code Recognition, Neural Computation, 1(4):541-551, 1989. See also Sec. 3 of [10].

[7] K. Yamaguchi, K. Sakamoto, A. Kenji, T. Akabane, Y. Fujimoto. A Neural Network for Speaker-Independent Isolated Word Recognition. First International Conference on Spoken Language Processing (ICSLP 90), Kobe, Japan, Nov 1990. A 1-dimensional convolutional TDNN using Max-Pooling instead of Fukushima's Spatial Averaging [1].

[8] Weng, J., Ahuja, N., and Huang, T. S. (1993). Learning recognition and segmentation of 3-D objects from 2-D images. Proc. 4th Intl. Conf. Computer Vision, Berlin, pp. 121-128. A 2-dimensional CNN whose downsampling layers use Max-Pooling (which has become very popular) instead of Fukushima's Spatial Averaging [1].

[9] In 2011, the fast and deep GPU-based CNN called DanNet (7+ layers) achieved the first superhuman performance in a computer vision contest. See overview: "2011: DanNet triggers deep CNN revolution."

[10] How 3 Turing awardees republished key methods and ideas whose creators they failed to credit. Technical Report IDSIA-23-23, Swiss AI Lab IDSIA, 14 Dec 2023. See also the YouTube video for the Bower Award Ceremony 2021: J. Schmidhuber lauds Kunihiko Fukushima.

[Nicolò Savioli, Ph.D. @NSavioli]

@drfeifei @theworldlabs It's just a dev job...

[Nicolò Savioli, Ph.D. @NSavioli]

Life is a stage of costumes and masks, where each person plays their part in an endless human drama.

[Nicolò Savioli, Ph.D. @NSavioli]

The brain isn’t purely digital or analog — it’s both. While quantum duality inspires speculation, it’s our biology that truly blends spikes and waves into thought.

[Nicolò Savioli, Ph.D. @NSavioli]

In other words:
the imagination of future pleasure guides us and makes our exploratory behavior more sensible, more intentional, and less random.

[Nicolò Savioli, Ph.D. @NSavioli]

Our exploration becomes more rational when guided by the foresight of the joy that lies ahead.

[Nicolò Savioli, Ph.D. @NSavioli]

@elonmusk @elonmusk Come to Italy to experience the beauty 😁

[Proxima @proximabio]

Announcing Neo-1: the world’s most advanced atomistic foundation model, unifying structure prediction and all-atom de novo generation for the first time - to decode and design the structure of life 🧵(1/10)

[Circuit @circuitrobotics]

The world’s first industrial superhumanoid robot?

Dexterity has released Mech - a superhumanoid, powered by Physical AI, to operate in complex, unstructured environments across industrial sites.

Unlike traditional robots, Mech moves with the skill of a human but has the strength and stamina of a machine, making it a powerful tool for tough industrial jobs.

As industries push for greater efficiency and safety, could superhumanoids like Mech redefine the future of industrial work?

[Nicolò Savioli, Ph.D. @NSavioli]

Human society can only achieve an evolutionary leap by reusing all information generated so far and reprocessing it through AI agents to uncover new latent knowledge. Like Homo sapiens, whose evolution is tied to the tools it creates, our future depends on intelligence that augments our own. #AI #Evolution #Future

[Nicolò Savioli, Ph.D. @NSavioli]

@Anoushkrit Diffusion is too slow...

[Nicolò Savioli, Ph.D. @NSavioli]

Transformers still dominate in 2025—parallel power & scale keep them king. But the field's stuck tweaking old tricks. Time for a real shake-up, not just faster RNNs or fancier attention. Where’s the next big leap?

[Nicolò Savioli, Ph.D. @NSavioli]

@pushmeet @nvidia @GoogleDeepMind @pushmeet Great partnership! But does this mean TPUs are officially in retirement? Guess it’s time to stock up on GPUs before they run out again! 😂

[Nicolò Savioli, Ph.D. @NSavioli]

@GiorgiaMeloni L’Italia potrebbe sostenere tra 150 e 300 AI factories utilizzando solo il 5-10% della sua energia rinnovabile. Questo ci renderebbe competitivi nell’AI senza aumentare il consumo di fonti fossili.

[Nicolò Savioli, Ph.D. @NSavioli]

@GiorgiaMeloni Italia ha 112,7 TWh di energia rinnovabile annua. Una GPU AI di fascia alta consuma fino a 700W, equivalenti a ~3,74 MWh/anno con utilizzo al 61%. Con 10.000 GPU, una AI factory richiederebbe 37,4 GWh/anno. Investire in AI factories sostenibili è possibile e strategico!

[Nicolò Savioli, Ph.D. @NSavioli]

@GiuseppeConteIT L’Italia potrebbe sostenere tra 150 e 300 AI factories utilizzando solo il 5-10% della sua energia rinnovabile. Questo ci renderebbe competitivi nell’AI senza aumentare il consumo di fonti fossili.

[Nicolò Savioli, Ph.D. @NSavioli]

@GiuseppeConteIT Italia ha 112,7 TWh di energia rinnovabile annua. Una GPU AI di fascia alta consuma fino a 700W, equivalenti a ~3,74 MWh/anno con utilizzo al 61%. Con 10.000 GPU, una AI factory richiederebbe 37,4 GWh/anno. Investire in AI factories sostenibili è possibile e strategico!

[Nicolò Savioli, Ph.D. @NSavioli]

@JaniceLama91342 @RaiaHadsell @TEDAIVienna Use a Bayesian network with dropout at test time.