From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mp11.migadu.com ([2001:41d0:8:6d80::]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)) by ms9.migadu.com with LMTPS id 0NDIInp9NmQFJQAASxT56A (envelope-from ) for ; Wed, 12 Apr 2023 11:44:26 +0200 Received: from aspmx1.migadu.com ([2001:41d0:8:6d80::]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)) by mp11.migadu.com with LMTPS id IJTJInp9NmR/TgEA9RJhRA (envelope-from ) for ; Wed, 12 Apr 2023 11:44:26 +0200 Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by aspmx1.migadu.com (Postfix) with ESMTPS id 527EBC464 for ; Wed, 12 Apr 2023 11:44:26 +0200 (CEST) Received: from localhost ([::1] helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1pmX1B-0005av-7t; Wed, 12 Apr 2023 05:43:57 -0400 Received: from eggs.gnu.org ([2001:470:142:3::10]) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1pmX17-0005aj-CI for guix-devel@gnu.org; Wed, 12 Apr 2023 05:43:54 -0400 Received: from mx0.riseup.net ([198.252.153.6]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1pmX14-0000vN-Tg for guix-devel@gnu.org; Wed, 12 Apr 2023 05:43:52 -0400 Received: from fews02-sea.riseup.net (unknown [10.0.1.112]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256 client-signature RSA-PSS (2048 bits) client-digest SHA256) (Client CN "mail.riseup.net", Issuer "R3" (not verified)) by mx0.riseup.net (Postfix) with ESMTPS id 4PxHpX0kPfz9sQf; Wed, 12 Apr 2023 09:43:48 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=riseup.net; s=squak; t=1681292628; bh=eJfWXDZBhXez3O2H+bBL9d/FoDZQ4tYeVrLqhtUdMOw=; h=References:From:To:Cc:Subject:Date:In-reply-to:From; b=QCTswAcGKUkpNiPwka9SlXzSIxAE+WYQRUXxxiihdg9l1gypw9tTcpfDk06EN91bD 1VWFnt9MonFtWDg7e+F9YEQ3Pfy2En+I5l5DMbvL8o0Ts364XsEf/ykoaddUHXon0w gFZlm/JpxNlivYGjx0icb3n1jr7SEFTRZXnfWGpY= X-Riseup-User-ID: 8FEF56653043583AFF25516300D4AB81112C3CF3943053083181DF571BF7C5A0 Received: from [127.0.0.1] (localhost [127.0.0.1]) by fews02-sea.riseup.net (Postfix) with ESMTPSA id 4PxHpW2TgnzFsRs; Wed, 12 Apr 2023 09:43:47 +0000 (UTC) References: <867cui6ci3.fsf@gmail.com> From: Csepp To: Nathan Dehnel Cc: Simon Tournier , rprior@protonmail.com, guix-devel@gnu.org Subject: Re: Guidelines for pre-trained ML model weight binaries (Was re: Where should we put machine learning model parameters?) Date: Wed, 12 Apr 2023 11:32:34 +0200 In-reply-to: Message-ID: <87sfd5qvub.fsf@riseup.net> MIME-Version: 1.0 Content-Type: text/plain Received-SPF: pass client-ip=198.252.153.6; envelope-from=raingloom@riseup.net; helo=mx0.riseup.net X-Spam_score_int: -27 X-Spam_score: -2.8 X-Spam_bar: -- X-Spam_report: (-2.8 / 5.0 requ) BAYES_00=-1.9, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_LOW=-0.7, SPF_HELO_PASS=-0.001, SPF_PASS=-0.001 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: guix-devel@gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: "Development of GNU Guix and the GNU System distribution." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: guix-devel-bounces+larch=yhetil.org@gnu.org Sender: guix-devel-bounces+larch=yhetil.org@gnu.org X-Migadu-Flow: FLOW_IN X-Migadu-Country: US ARC-Seal: i=1; s=key1; d=yhetil.org; t=1681292666; a=rsa-sha256; cv=none; b=ocSsuIl3V/wO3YLFo3z6leRvogHE7m1VH8wy47h6yfXydsW8tUymMmkSNjw//H7r/ebgs2 cOecA9aRreThABfACdClKdvgsz1eTeKqyZeaVJ0HPD7OvhXAKsii8dw2H8MAModqyFQzRZ fV/SZgEA3QoENvkbXrgVPVIk3x/Rn3XlGO252Ij3b/pib6LDEaX/REXc2s4VrNS3IRlo9U U2Jr4eT/oebMxJXymawq1NON5sh9gl6OYwFbOhYmTMaJyDYXntGt0Btz6uAadfY2g9AqQq ke/TGnrDHww3nH6XpyVufSQinmctW4YIcQR+Yi0qZjttM0jLjXDowaq/gKiPWQ== ARC-Authentication-Results: i=1; aspmx1.migadu.com; dkim=pass header.d=riseup.net header.s=squak header.b=QCTswAcG; spf=pass (aspmx1.migadu.com: domain of "guix-devel-bounces+larch=yhetil.org@gnu.org" designates 209.51.188.17 as permitted sender) smtp.mailfrom="guix-devel-bounces+larch=yhetil.org@gnu.org"; dmarc=pass (policy=none) header.from=riseup.net ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=yhetil.org; s=key1; t=1681292666; h=from:from:sender:sender:reply-to:subject:subject:date:date: message-id:message-id:to:to:cc:cc:mime-version:mime-version: content-type:content-type:in-reply-to:in-reply-to: references:references:list-id:list-help:list-unsubscribe: list-subscribe:list-post:dkim-signature; bh=HA1jvankyFsO9G0Ng8MhVPYJO95CeksOKEAY9+mP/sI=; b=bzyddw3UmEu5abfSR5+cPOOmonMQ+smvffFghmcJZerXRBxurWGK1AziT/UIAfthek4eIS e560W1J5ZEBKcOQOf1mP0oOcmf8KlZBad1FGizJsROzpbTpzp0xpAXMMd7Dvrv9tswnhs9 nLl1IUsLquN3DWeByal/ET0p9+WgT7hbMKoB9LknwzgOnbXtgV5HjrOWUQsn3aiBaoXJjd IaefZKdZq/y//nJA3Vwjkpv4s0Pxr14GC3+HvejbbJ35R3/pjvyAThGwnVibk2IRKcN853 J/rVmWUHxk07p09ABS4GRm6j+id6ZqdhkuERucm0q/WBKHGpV5I3Bsk06xrACQ== X-Migadu-Queue-Id: 527EBC464 Authentication-Results: aspmx1.migadu.com; dkim=pass header.d=riseup.net header.s=squak header.b=QCTswAcG; spf=pass (aspmx1.migadu.com: domain of "guix-devel-bounces+larch=yhetil.org@gnu.org" designates 209.51.188.17 as permitted sender) smtp.mailfrom="guix-devel-bounces+larch=yhetil.org@gnu.org"; dmarc=pass (policy=none) header.from=riseup.net X-Migadu-Scanner: scn0.migadu.com X-Migadu-Spam-Score: -1.60 X-Spam-Score: -1.60 X-TUID: O70mN6eFjhox Nathan Dehnel writes: > a) Bit-identical re-train of ML models is similar to #2; other said > that bit-identical re-training of ML model weights does not protect > much against biased training. The only protection against biased > training is by human expertise. > > Yeah, I didn't mean to give the impression that I thought > bit-reproducibility was the silver bullet for AI backdoors with that > analogy. I guess my argument is this: if they release the training > info, either 1) it does not produce the bias/backdoor of the trained > model, so there's no problem, or 2) it does, in which case an expert > will be able to look at it and go "wait, that's not right", and will > raise an alarm, and it will go public. The expert does not need to be > affiliated with guix, but guix will eventually hear about it. Similar > to how a normal security vulnerability works. > > b) The resources (human, financial, hardware, etc.) for re-training is, > for most of the cases, not affordable. Not because it would be > difficult or because the task is complex, this is covered by the > point a), no it is because the requirements in term of resources is > just to high. > > Maybe distributed substitutes could change that equation? Probably not, it would require distributed *builds*. Right now Guix can't even use distcc, so it definitely can't use remote GPUs.