Un team di ricercatori di Hong Kong ha reso pubblico un sistema denominato Lancelot, che rappresenta la prima realizzazione pratica di apprendimento federato, risultando al contempo protetto da attacchi di manomissione dei dati e da violazioni della riservatezza.

L’apprendimento federato consente a più partecipanti (client) di addestrare congiuntamente un modello senza rivelare i dati di origine. Questo approccio è particolarmente importante in medicina e finanza, dove le informazioni personali sono strettamente regolamentate.

Tuttavia, questi sistemi sono vulnerabili al data poisoning : un aggressore può caricare aggiornamenti falsi e distorcere i risultati. I metodi di apprendimento federato hanno parzialmente risolto questo problema scartando gli aggiornamenti sospetti, ma non hanno protetto dal possibile recupero di dati crittografati dalla memoria della rete neurale.

Il team ha deciso di combinare sicurezza crittografica e resistenza agli attacchi. Lancelot utilizza la crittografia completamente omomorfica per garantire che tutti gli aggiornamenti del modello locale rimangano crittografati end-to-end.

Il sistema seleziona inoltre gli aggiornamenti client attendibili senza rivelare chi è attendibile. Ciò è possibile grazie a uno speciale meccanismo di “ordinamento mascherato”: un centro chiavi attendibile riceve i dati crittografati, ordina i client in base al livello di attendibilità e restituisce al server solo un elenco crittografato, oscurando i partecipanti alla formazione. In questo modo, il server aggrega solo i dati verificati senza rivelarne l’origine.

Per velocizzare i calcoli, gli sviluppatori hanno implementato due tecniche di ottimizzazione. La “rilinearizzazione lazy” posticipa i costosi passaggi crittografici alla fase finale, riducendo il carico sulla CPU. Il metodo “Dynamic hoisting” raggruppa le operazioni ripetitive e le esegue in parallelo, anche sulle GPU, riducendo significativamente i tempi di addestramento complessivi.

Il risultato è una soluzione che affronta due vulnerabilità dell’apprendimento federato: è resiliente agli attacchi di malintenzionati e al contempo garantisce la completa riservatezza dei dati. I test hanno dimostrato che Lancelot non solo previene fughe di dati e sabotaggi, ma riduce anche significativamente i tempi di addestramento dei modelli ottimizzando le operazioni crittografiche e sfruttando le GPU.

I ricercatori intendono espandere l’architettura Lancelot, rendendola adatta a scenari su larga scala. Le potenziali applicazioni includono l’addestramento di sistemi di intelligenza artificiale in ospedali, banche e altre organizzazioni che gestiscono dati sensibili. Il team sta attualmente testando nuove versioni con supporto per chiavi distribuite (CKKS a soglia e multi-chiave), integrazione di metodi di privacy differenziale e aggregazione asincrona, che consentiranno al sistema di funzionare in modo affidabile anche con connessioni di rete instabili e un’ampia varietà di dispositivi client.

L'articolo Lancelot: il sistema di apprendimento AI federato e sicuro proviene da Red Hot Cyber.

The physical layout of the SCHEME-78 LISP-based microprocessor by Steele and Sussman. (Source: ACM, Vol 23, Issue 11, 1980)
During the AI research boom of the 1970s, the LISP language – from LISt Processor – saw a major surge in use and development, including many dialects being developed. One of these dialects was Scheme, developed by [Guy L. Steele] and [Gerald Jay Sussman], who wrote a number of articles that were published by the Massachusetts Institute of Technology (MIT) AI Lab as part of the AI Memos. This subset, called the Lambda Papers, cover the ideas from both men about lambda calculus, its application with LISP and ultimately the 1980 paper on the design of a LISP-based microprocessor.

Scheme is notable here because it influenced the development of what would be standardized in 1994 as Common Lisp, which is what can be called ‘modern Lisp’. The idea of creating dedicated LISP machines was not a new one, driven by the processing requirements of AI systems. The mismatch between the S-expressions of LISP and the typical way that assembly uses the CPUs of the era led to the development of CPUs with dedicated hardware support for LISP.

The design described by [Steele] and [Sussman] in their 1980 paper, as featured in the Communications of the ACM, features an instruction set architecture (ISA) that matches the LISP language more closely. As described, it is effectively a hardware-based LISP interpreter, implemented in a VLSI chip, called the SCHEME-78. By moving as much as possible into hardware, obviously performance is much improved. This is somewhat like how today’s AI boom is based around dedicated vector processors that excel at inference, unlike generic CPUs.

During the 1980s LISP machines began to integrate more and more hardware features, with the Symbolics and LMI systems featuring heavily. Later these systems also began to be marketed towards non-AI uses like 3D modelling and computer graphics. As however funding for AI research dried up and commodity hardware began to outpace specialized processors, so too did these systems vanish.

Top image: Symbolics 3620 and LMI Lambda Lisp machines (Credit: Jason Riedy)

hackaday.com/2025/10/20/the-la…

Testing the FOC-based motor controller. (Credit: Excessive Overkill, YouTube)
Vector Control, also known as Field Oriented Control or FOC is an AC motor control scheme that enables fine-grained control over a connected motor, through the precise control of its phases. In a recent video [Excessive Overkill] goes through the basics and then the finer details of how FOC works, as well as how to implement it. These controllers generally uses a proportional integral (PI) loop, capable of measuring and integrating the position of the connected motor, thus allowing for precise adjustments of the applied vector.

If this controller looks familiar, it is because we featured it previously in the context of reviving old industrial robotic arms. Whether you are driving the big motors on an industrial robot, or a much smaller permanent magnet AC (PMAC) motor, FOV is very likely the control mechanism that you want to use for the best results. Of note is that most BLDC motors are actually also PMACs with ESC to provide a DC interface.

The actual driving is done with two MOSFETs per phase, forming a half-bridge, switching between the two rails to create the requisite PWM signal for each phase. Picking the right type of MOSFET was somewhat hard, especially due to the high switching currents and the high frequency at 25 kHz. The latter was picked to prevent audible noise while driving a robot. Ultimately SiC MOSFETs were picked, specially the GeneSiC G3R30MT12K. Of note here are the four legs, with a fourth Kelvin Source pin added. This is to deal with potential gate drive issues that are explained in the video.

With the hardware in place, whether following the [Excessive Overkill] GitHub projects or not, what makes all of it work is the software. This is where the microcontroller aspect is essential, as it has to do all the heavy lifting of calculating the new optimal vector and thus the current levels per phase. In this controller an STM32F413 is used, which generates the PWM signals to drive the half-bridges, while reading the measurements from the motors with its ADC.

As can be seen in the resulting use of this controller with old industrial robots, the FOC controller works quite well, with quiet and smooth operation. This performance is why we’re likely to see FOC and PMAC motors used in applications like 3D printers in the future, though the rule of ‘good enough’ makes the cost of an FOC controller still a tough upsell over a simple open loop stepper-based system.

youtube.com/embed/ujofKWmGChw?…

hackaday.com/2025/10/20/high-p…

Let’s say you want to blink an LED. You might grab an Arduino and run the Blink sketch, or you might lace up a few components to a 555. But you needn’t go so fancy! [The Design Graveyard] explains how this same effect can be achieved with a single transistor.

The circuit in question is rather odd at first blush. The BC547 NPN transistor is hooked up between an LED and a resistor leading to a 12V DC line, with a capacitor across the emitter and collector. Meanwhile, the base is connected to… nothing! It’s just free-floating in the universe of its own accord. You might expect this circuit to do nothing at all, but if you power it up, the LED will actually start to flash.

The mechanism at play is relatively simple. The capacitor charges to 12 volts via the resistor. At this point, the transistor, which is effectively just acting as a poor diode in this case, undergoes avalanche breakdown at about 8.5 to 9 volts, and starts conducting. This causes the capacitor to discharge via the LED, until the voltage gets low enough that the transistor stops conducting once again. Then, the capacitor begins to charge back up, and the cycle begins again.

It’s a weird way to flash an LED, and it’s not really the normal way to use a transistor—you’re very much running it out of spec. Regardless, it does work for a time! We’ve looked at similar circuits before too. Video after the break.

youtube.com/embed/Yw6L5w4TDxw?…

[Thanks to Vik Olliver for the tip!]

hackaday.com/2025/10/20/blinki…

Making nylon plastic from raw chemicals used to be a very common demo; depending where and when you grew up, you may well have done it in high school or even earlier. What’s not common is taking that nylon and doing something with it, like, say extruding it into filament to make a benchy. [Startup Chuck] shows us there might be a reason for that. (Video, embedded below.)

It starts out well enough: sebacoyl chloride and hexamethaline diamine mix up and do their polymerizing tango to make some nylon, just like we remember. (Some of us also got to play with mercury bare-handed; safety standards have changed and you’ll want to be very careful if you try this reaction at home). The string of nylon [Chuck] pulls from the beaker even looks a little bit like filament for a second, at least until it breaks and gets tossed into a blobby mess. We wonder if it would be possible to pull nylon directly into 1.75 mm filament with the proper technique, but quality control would be a big issue. Even if you could get a consistent diameter, there’d likely be too much solvent trapped inside to safely print.

Of course, melting the nylon with a blowtorch and trying to manually push the liquid through a die to create filament has its own quality control problems. That’s actually where this ends: no filament, and definitely no benchy. [Chuck] leaves the challenge open to anyone else who wants to take the crown. Perhaps one of you can show him how it’s done. We suspect it would be easiest to dry the homemade nylon and shred it into granules and only then extrude them, like was done with polypropylene in this mask-recycling project. Making filament from granules or pellets is something we’ve seen more than once over the years.

If you really want to make plastic from scratch, ordering monomers from Sigma-Aldrich might not cut it for ultimate bragging rights; other people are starting with pulling CO2 from the atmosphere.

Thanks to [Chaz] for the tip! Remember that the tips line isn’t just for your successes– anything interesting can find its home here.

youtube.com/embed/0DwcGJUcSyE?…

hackaday.com/2025/10/20/fail-o…

What does psychosocial support look like in the face of spyware attacks and digital security threats? It can mean adapting care to the context, listening without rushing, and building protocols that protect both dignity and data. But we want to hear what it means to you and to those already integrating psychosocial support into their accompaniments — such as Fundación Acceso in Latin America and Digital Society of Africa during our next community call.

The post Community Call: Psychosocial Support & Digital Safety appeared first on European Digital Rights (EDRi).

Learn more about the state of digital attacks on Human Rights Defenders, when researchers and experts come together to discuss a recently concluded report.

The post DDI Knowledge Hub Community Call: On mapping of HRD support mechanisms with Expectation State appeared first on European Digital Rights (EDRi).

During this lab, you will establish a strong foundation in designing participatory and interactive sessions for both online and face-to-face formats. This will enable you to facilitate sessions that foster co-empowering, learning, and knowledge sharing, thereby advancing your work and that of your participants.

The post The Session Design Lab appeared first on European Digital Rights (EDRi).

As national and regional efforts around digital sovereignty gain momentum, SplinterCon Paris will bring together technologists, researchers, and policymakers to consider what resilient, interconnected digital societies require today and in the future.

The post SplinterCon in Paris appeared first on European Digital Rights (EDRi).

If you live in Cambridge, work in Cambridge, travel through Cambridge or live in the Greater Cambridge-area, please tell the Cambridge City Council that you oppose the introduction of Flock ALPR surveillance cameras there. We talk about it in the latest Pirate News, but you can find out more especially about today’s meeting at 5:30pm at our post about it.

Joe, Steve and James discuss why we oppose Cambridge’s effort to install Flock ALPR cameras, what we are doing at the Boston Anarchist Bookfair, the Boston NoKings protest and CycloMedia.

youtube.com/embed/9zGsUH5n9QI?…

Join us at the Boston Anarchist Bookfair Nov. 1-2.

Sign up to our newsletter to get notified of new events or volunteer. Join us on:

• Mastodon;
• Facebook;
• Blue Sky;
• Twitter.

Check out:

• Mon. & Thu., Speak Out Against Cambridge ALPR Surveillance Rollout;
• Surveillance Memory Bank and Resources;
• Our Administrative Coup Memory Bank;
• Our Things to Do when Fascists are Taking Over.

Some links we mentioned:

• Boston Anarchist Bookfair;
• Steve’s Bluesky post about Boston Lobsters at the Boston NoKings rally;
• Steve’s Bluesky post about CycloMedia mapping SUV;
• CycloMedia Wikipedia page.

Image Credit: James O’Keefe, CC By-SA 4.0

masspirates.org/blog/2025/10/2…

Dopo circa sedici anni di Ubuntu come sistema operativo sul mio PC personale, ho finalmente fatto distro hopping verso #debian (Debian 13 Trixie).

L'ho presa molto comoda, ma finalmente un passo in avanti verso una distro più "comunitaria" e politiche di scelta programmi meno invadenti e vincolanti per l'utente.

Siamo all'Alessandrino in via del Grano alle ore 18.00 di oggi. Guardate il risultato della raccolta stradale con i cassonetti. A Roma non differenziano: nei cassonetti per l'indifferenziato sversano di tutto, compresi tanti materiali facilmente riciclabili.
A Roma, insomma, voto zero in raccolta differenziata.
Il sostegno a Gualtieri nella gestione del ciclo dei rifiuti equivale a essere partecipi di un fallimento, complici di un crimine ambientale e corresponsabili per danno erariale.