Gold to Reach $12,000? Arthur Hayes Deciphers the 'War Code' Behind It

比推Опубліковано о 2026-01-15Востаннє оновлено о 2026-01-15

Анотація

Arthur Hayes argues that Bitcoin's underperformance in 2025, compared to gold and the Nasdaq, is due to a decline in US dollar liquidity. He posits that gold's surge is driven by price-insensitive central bank buying, fueled by a global shift away from US Treasury holdings following the 2022 seizure of Russian assets. This move towards settling trade deficits with gold supports its rise independent of dollar liquidity. Conversely, the Nasdaq's resilience is attributed to the "nationalization" of AI by the US and China, where government-backed capital expenditure continues to fuel growth of strategic tech companies despite tighter liquidity. Hayes predicts a significant expansion of dollar liquidity in 2026 through Federal Reserve actions, increased bank lending, and lower mortgage rates, which he believes will propel Bitcoin's price upward. His trading strategy involves leveraged Bitcoin exposure through stocks like MicroStrategy and Metaplanet.

Author:Arthur Hayes

Title:Frowny Cloud

Compiled and Edited by:BitPushNews


(Disclaimer: All views expressed herein are solely the author's personal opinions and should not be the basis for investment decisions, nor should they be considered advice or recommendations for engaging in investment transactions.)

The deities I worship have all taken the form of adorable plush toys.

During the peak ski season in Hokkaido in January and February, I pray to the "Frowny Cloud"—the little deity who governs snowfall. The local climate pattern dictates that during the best part of the season, snow falls almost day and night, and you hardly see the sun. Fortunately, I also pray to the Vitamin God—a soft, cute little horse plush toy

—who bestows upon me Vitamin D3 tablets and various other blessings.

Although I love snow, not all snow is high-quality and safe. The kind of carefree, aggressive skiing experience I enjoy requires a specific type of snow: low wind speeds at night, with temperatures between -5 and -10 degrees Celsius. Under these conditions, new snow effectively bonds with the old snow, creating bottomless powder. During the day, the Frowny Cloud blocks specific wavelengths of sunlight, preventing south-facing slopes and others from being "sun-cooked," which could lead to potential avalanches.

Sometimes, Frowny Cloud abandons us fearless skiers at night. Cold, clear nights cause the snow layers to develop "layer separation" after warming and cooling, forming persistent weak layers. This phenomenon can exist in the snowpack for a long time, and once triggered by a skier's weight causing energy transfer and collapse, it can lead to deadly avalanches.

As always, the only way to understand what kind of snow layers Frowny Cloud has created is to study history. On the slopes, we do this by digging huge pits and analyzing the different types of snow that have fallen over time. But since this isn't an article about avalanche theory, our approach in the market is to study charts and the interaction between historical events and price fluctuations.

In this article, I hope to examine the relationship between Bitcoin, gold, stocks (especially the US tech giants in the Nasdaq 100 index), and US dollar liquidity. Those who are Gold Bugs or members of the financial establishment adorned with Hermès scarves and red-soled shoes (who firmly believe in "long-term stock holding"—I didn't get good enough grades at Wharton to get into Professor Siegel's class) are thrilled that Bitcoin has been the worst-performing mainstream asset in 2025.

These Gold Bugs sneer at Bitcoin supporters: if Bitcoin is touted as a vote against the established order, why hasn't its performance matched or exceeded that of gold? The dirty fiat currency stock peddlers also scoff: Bitcoin is just a "high-beta" (high-risk) plaything of the Nasdaq, but in 2025 it didn't even keep up, so why should cryptocurrency be considered in asset allocation?

This article will present a series of beautiful charts, accompanied by my annotations, to clarify the联动关系 (linkage relationships) of these assets.

I believe Bitcoin's performance is entirely as expected.

It rides the waves of fiat currency liquidity—especially US dollar liquidity, because the credit impulse of "Pax Americana" is the most important force in 2025.

Gold's surge is because price-insensitive sovereign nations are hoarding it frantically, afraid that their wealth will be confiscated by the US if left in US Treasuries (as happened to Russia in 2022).

Recent US actions against Venezuela will only further strengthen countries' desire to hold gold instead of US Treasuries. Finally, the AI bubble and its related industries are not going away. In fact, Trump must double down on state support for AI because it is the largest contributor to the empire's GDP growth. This means that even if the pace of dollar creation slows, the Nasdaq can continue to rise because Trump has effectively "nationalized" it.

If you've studied China's capital markets, you know that stocks perform very well in the early stages of nationalization but then significantly underperform as political goals take precedence over capitalists' returns.

If the price movements of Bitcoin, gold, and stocks in 2025 validate my market structure, then I can continue to focus on the ebbs and flows of US dollar liquidity.

Reminding readers, my prediction is: Trump will疯狂注入信贷 (crazily inject credit) to make the economy run "hot to the point of explosion." A booming economy helps the Republican Party win in this November's election. As central bank balance sheets expand, commercial banks increase loans to "strategic industries," and mortgage interest rates fall due to money printing, US dollar credit will expand significantly.

In summary, does this mean I can continue to "surf" carefree—that is, aggressively deploy the fiat currency I earn and maintain maximum risk exposure? I leave it to the readers to judge.

The Chart That Rules Them All

First, let's compare the returns of Bitcoin, gold, and the Nasdaq in the first year of Trump's second term. How do these assets' performances compare to changes in US dollar liquidity?

I will elaborate later, but the basic assumption is: if US dollar liquidity declines, these assets should also fall. However, gold and stocks have risen. Bitcoin performed as expected: like dog shit. Next, I will explain why gold and stocks have managed to rise against the trend despite falling US dollar liquidity.

[Chart: Bitcoin (red), Gold (gold), Nasdaq 100 (green), and US Dollar Liquidity (purple) Comparison]

All That Glitters Is Not Gold, But Gold Is Indeed Glittering

My cryptocurrency journey began with gold. In 2010 and 2011, as the Fed-led quantitative easing (QE) intensified, I started buying physical gold in Hong Kong. Although the absolute amount was pitifully small, it accounted for a surprisingly high percentage of my net worth at the time.

Eventually, I learned a painful lesson about position management because I had to sell gold at a loss to buy Bitcoin for arbitrage in 2013. Fortunately, the ending was happy. Even so, I still hold significant amounts of physical gold coins and bars in vaults around the world, and my stock portfolio is dominated by gold and silver mining stocks. Readers might wonder: since I am a devoted believer in Satoshi Nakamoto, why do I still hold gold?

I hold gold because we are in the early stages of global central banks selling US Treasuries and buying gold. Additionally, countries are increasingly using gold to settle trade deficits, even when analyzing the US trade deficit.

In short, I buy gold because central banks are buying it. Gold, as civilization's true货币 (currency), has a 10,000-year history. Therefore, no significant central bank reserve manager would存 (store) Bitcoin when distrusting the current dollar-dominated financial system; they would and are buying gold. If gold's share of total global central bank reserves returns to the level of the 1980s, the gold price will rise to $12,000. Before you think I'm dreaming, let me prove it to you intuitively.

In the fiat system, the traditional view of gold is as an inflation hedge. Therefore, it should roughly track the empire-manipulated CPI index. The chart above shows that since the 1930s, gold has roughly tracked this index. However, starting in 2008 and accelerating after 2022, the gold price has risen far faster than inflation. So, is gold in a bubble, ready to收割 (harvest) gamblers like me?

[Chart: Gold Price vs. US CPI]

If gold were in a bubble, retail investors should be flocking to it. The most popular way to trade gold is through ETFs, with GLD being the largest. When retail investors疯狂买入 (frantically buy) gold, the number of GLD shares outstanding increases. To compare across different periods and gold price regimes, we must divide the number of GLD shares outstanding by the physical gold price. The chart below shows that this ratio is falling, not rising, meaning the true gold speculation frenzy has not yet arrived.

[Chart: GLD Shares Outstanding Divided by Gold Spot Price]

If it's not retail investors pushing up the gold price, who are those price-insensitive buyers? They are central bank governors around the world. Over the past two decades, there have been two key moments when these people realized: US money is only good for wiping your ass.

In 2008, US financial titans created a global deflationary financial crisis. Unlike in 1929 when the Fed did not intervene, this time the Fed violated its obligation to maintain the dollar's purchasing power and printed money like crazy to "save" specific large financial players. This marked a watershed in the proportion of sovereign holdings of US Treasuries and gold.

In 2022, President Biden shocked the world by freezing the Treasury holdings of a country with a vast nuclear arsenal and being the world's largest exporter of大宗商品 (commodities) (Russia). If the US is willing to abolish Russia's property rights, it can do the same to any weaker or less resource-rich country. Not surprisingly, other countries can no longer comfortably increase their exposure to US Treasuries that face the risk of confiscation. They began to accelerate gold purchases. Central banks are price-insensitive buyers. If the US President steals your money, your assets instantly become zero. Since buying gold eliminates counterparty risk, what does it matter if the price is a bit expensive?

The most fundamental reason for sovereign nations' infinite appetite for this "barbarous relic" is: net trade settlement is increasingly conducted through gold. The record shrinkage of the US trade deficit in December 2025 is evidence of gold re-establishing itself as the global reserve currency. Over 100% of the change in the US net trade balance was attributable to gold exports.

"The goods import-export gap fell 11% from the previous month to $52.8 billion, according to data released by the Commerce Department on Thursday. This brought the deficit to its smallest level since June 2020...... Exports rose 3% in August to $289.3 billion, driven mainly by non-monetary gold." — Source: Financial Times

The flow path of gold is: the US exports gold to Switzerland, where it is refined and recast, then shipped to other countries. The chart below shows that it is mainly China, India, and other emerging economies that manufacture physical goods or export commodities that buy this gold. The physical goods ultimately flow to the US, while the gold flows to the more productive regions of the world.

By "productive," I don't mean these places are better at writing bullshit reports or annotating complex email signatures, but that they export energy and other key industrial commodities, and their people produce steel and refine rare earths. Gold rises despite falling US dollar liquidity because sovereign nations are accelerating the restoration of the global gold standard.

[Chart: Gold Import/Export Flow Map by Country]

Long-Termists Love Liquidity

Every era has its high-flying tech stocks. In the roaring US bull market of the 1920s, radio manufacturer RCA was the hot tech darling; in the 1960s and 1970s, IBM, maker of newfangled mainframe computers, was the market focus; today, AI hyperscalers and chip makers are all the rage.

Humans are inherently optimistic. We love to predict a glorious future: every dollar spent by tech companies today will bring about a social utopia tomorrow. To realize this vision in investors' minds, companies burn cash and take on debt. When liquidity is cheap, betting on the future becomes easy. Therefore, investors are happy to spend cheap cash today on tech stocks in exchange for the chance of huge cash flows in the future, driving up P/E ratios. So, during periods of excess liquidity, tech growth stocks rise exponentially.

Bitcoin is monetary technology. The value of this technology is only relative to the degree of fiat currency depreciation. The creation of Proof-of-Work (PoW) blockchain is great, which itself guarantees that Bitcoin's value is greater than zero. But for Bitcoin's value to approach $100,000, sustained fiat货币贬值 (currency depreciation) is needed. Bitcoin's asynchronous growth is a direct result of the explosive growth in the US money supply after the 2008 global financial crisis.

Therefore, I say: When US dollar liquidity expands, Bitcoin and the Nasdaq rise.

The only contradiction at present is: the recent divergence between Bitcoin's price and the Nasdaq.

[Chart: Bitcoin vs. Nasdaq Price Trend]

My theory for why the Nasdaq did not correct with the decline in US dollar liquidity in 2025 is: AI has been "nationalized" by both China and the US.

AI tech moguls have sold the world's two great leaders on the idea that AI can solve everything. AI can reduce labor costs to zero, cure cancer, increase productivity, and most importantly, achieve military dominance over the globe. Therefore, whichever country "wins" in AI rules the world. China bought into this long ago; it fits perfectly with its five-year plans.

In the US, this analysis is new, but industrial policy is just as entrenched as in China, only marketed differently. Trump drank the AI "Kool-Aid," and "winning the AI race" became his economic platform. The US government has effectively nationalized any component deemed helpful to "winning." Through executive orders and government investment, Trump is blunting market signals and causing capital to flood into AI-related areas regardless of returns. This is why the Nasdaq decoupled from Bitcoin and the decline in US dollar liquidity in 2025.

[Chart: Nasdaq vs. US Dollar Liquidity Decoupling Chart]

Bubble or not, increased spending to "win" AI is driving the US economy. Trump promised to make the economy run hot; he can't stop just because the return on this spending might be below the cost of capital a few years later.

US tech investors should be cautious. Industrial policy aimed at "winning AI" is an excellent way to burn money. Trump's (or his successor's) political goals will diverge from the interests of shareholders in strategic enterprises. This is a lesson Chinese stock investors learned the hard way. As Confucius said, "Study the past to know the future." Clearly, given the Nasdaq's outstanding performance, this lesson hasn't been learned by US investors yet.

[Chart: US PMI and Economic Growth Data]

PMI readings below 50 indicate contraction. All the GDP growth hasn't brought a manufacturing renaissance. I thought Trump was for the white blue-collar workers? No, bro. Clinton sold your jobs to China, Trump brought the factories back, but now the factory floors are full of AI robot arms owned by Musk. Sorry, you got played again!

These charts clearly show that the Nasdaq's rise is US government-supported. Therefore, even if overall US dollar credit growth is weak, the AI industry will get all the capital it needs to win. The Nasdaq has thus decoupled and outperformed Bitcoin. I don't think the AI bubble is ready to burst. This outperformance will remain a feature of global capital markets until it isn't, or most likely until the red team loses the House in 2026 (as predicted by Polymarket). If the Republicans are The Jetsons (techies), then the Democrats are The Flintstones (retro).

If both gold and the Nasdaq have momentum, how can Bitcoin regain its footing? US dollar liquidity must expand. Clearly, I believe this will happen in 2026; let's explore how.

Make the Economy Hot to the Point of Explosion

At the beginning, I said there are three pillars supporting the dramatic increase in US dollar liquidity this year:

  1. The Fed's balance sheet will expand due to money printing.

  2. Commercial banks will lend to strategic industries.

  3. Mortgage interest rates will fall due to money printing.

[Chart: Fed Balance Sheet Size]

The Fed's balance sheet declined in 2025 due to quantitative tightening (QT). QT ended in December, and at that month's meeting, a new money-printing program called "Reserve Management Purchases" (RMP) was launched. I discussed this in depth in a previous article. The chart clearly shows the balance sheet bottomed in December. RMP injects at least $40 billion per month, and its size will increase as the funding needed to finance the US government grows.

[Chart: US Banking Loan Growth (ODL)]

The chart above is the banking loan growth indicator (ODL) released by the Fed. Starting from Q4 2025, banks issued more loans. When banks issue loans, they create deposits out of thin air, thus creating money. Banks like JPMorgan are very happy to lend to government-backed enterprises. JPMorgan launched a $1.5 trillion lending program for this. The process is: the government invests in an enterprise, the bank sees the government backing reduces default risk, and is happy to create money to fund this strategic industry. This is exactly what China has been doing. Credit creation shifts from the central bank to the commercial banking system, with a higher money multiplier initially, creating超趋势 (above-trend) nominal GDP growth.

The US will continue to display military force, and producing weapons of mass destruction requires financing from the commercial banking system. This is why bank credit growth will see a structural rise in 2026.

Trump is a real estate businessman; he knows how to finance property. His new order is for Freddie Mac and Fannie Mae (the "GSEs") to use the capital on their balance sheets to buy $200 billion in mortgage-backed securities (MBS). This is a net increase in US dollar liquidity. If successful, Trump won't stop there. Boosting the housing market by lowering mortgage rates will allow Americans to take out home equity loans. This wealth effect will put voters in a good mood on election day,转而支持 (turning to support) the Republicans. More importantly, it creates more credit to buy financial assets.

[Chart: Bitcoin and US Dollar Liquidity Bottom Coincidence Chart]

Bitcoin and US dollar liquidity bottomed almost simultaneously. As US dollar liquidity increases rapidly for the reasons above, Bitcoin will set sail. Forget about the 2025 performance; that was due to insufficient liquidity.

Trading Strategy

I am an aggressive speculator. Although my fund, Maelstrom, is almost fully invested, I still want to increase risk exposure because I am extremely bullish on US dollar liquidity growth. Therefore, I am gaining leveraged exposure to Bitcoin by going long on MicroStrategy (MSTR) and Metaplanet (3350 JT), without trading complex options or perpetual contracts.

[Chart: MSTR and Metaplanet Price Ratio Relative to Bitcoin]

I divided the share prices of these two companies by the Bitcoin price. They are currently at the low end of their range over the past two years. If Bitcoin can reclaim $110,000, investors will get the urge to go long Bitcoin through these instruments. Given the leverage embedded in these companies' balance sheets, they will far outperform Bitcoin in an uptrend.

Additionally, we continue to accumulate Zcash (ZEC). The departure of ECC developers is not bearish; I believe they can deliver more impactful products in their own for-profit entity. I am grateful for the opportunity to buy discounted ZEC from "weak hands."

Onward, crypto adventurers. The world is full of dangers; please remain vigilant. May peace be with you—and, pray to the "Frowny Cloud."


Original link:https://www.bitpush.news/articles/7602898

Пов'язані питання

QAccording to Arthur Hayes, why has gold been performing well despite a decline in dollar liquidity?

AGold has been performing well because price-insensitive sovereign nations are aggressively hoarding it, driven by fears of wealth confiscation in US Treasuries (as happened to Russia in 2022) and an increasing use of gold to settle trade deficits, effectively moving towards a de facto global gold standard.

QWhat is Hayes' prediction for the price of gold if it returns to its 1980s proportion of global central bank reserves?

AHayes predicts that if gold returns to its 1980s proportion of global central bank reserves, the price will reach $12,000 per ounce.

QHow does Hayes explain the recent decoupling of the Nasdaq's performance from Bitcoin and dollar liquidity?

AHayes explains that the Nasdaq has decoupled because AI has been effectively 'nationalized' by the US and China. Government support and industrial policy, aimed at 'winning' the AI race, are directing capital into AI-related sectors regardless of market signals or return on capital, insulating it from broader dollar liquidity trends.

QWhat are the three pillars Hayes identifies that will lead to a dramatic expansion of US dollar liquidity in 2026?

AThe three pillars are: 1) Expansion of the Fed's balance sheet through money printing (RMP program), 2) Increased lending by commercial banks to 'strategic industries' with government backing, and 3) A decrease in mortgage rates due to money printing, facilitated by agencies like Fannie Mae and Freddie Mac buying MBS.

QWhat trading strategy does Hayes employ to gain leveraged exposure to Bitcoin's expected price increase?

AHayes gains leveraged exposure to Bitcoin by going long on MicroStrategy (MSTR) and Metaplanet (3350 JT) stock. These companies trade at a low ratio to Bitcoin's price and have leverage embedded in their balance sheets, meaning they are expected to outperform Bitcoin significantly during a price rally.

Пов'язані матеріали

Former CFTC Chairman, Circle President Tarbert: Preaching Long-Termism While Cashing Out $30 Million Himself

Former CFTC Chairman and Circle President Heath Tarbert has consistently advocated for a long-term vision in public, urging patience from investors as Circle’s stock price has fallen significantly from its peak. However, it has been revealed that since Circle’s IPO, Tarbert has continuously sold his CRCL shares through pre-arranged trading plans, cashing out approximately $30 million, without making any public market purchases. This contrast between his public messaging and personal actions has drawn criticism. Tarbert joined Circle in July 2023 as Chief Legal Officer, leveraging his regulatory experience to help guide the company through its IPO and expansion. Despite promoting stablecoins as long-term infrastructure, he established a 10b5-1 trading plan just before Circle went public, leading to substantial stock sales over the following year. In March 2026, he initiated another plan to sell more shares. His career trajectory highlights a pattern of moving between high-level regulatory roles and influential positions in the financial sector. After resigning as CFTC Chairman in early 2021, he joined Citadel Securities as Chief Legal Officer just 27 days later, during a period of intense regulatory scrutiny for the firm. He later joined Circle, aiding its efforts to navigate regulatory challenges for its public listing. While Tarbert's expertise in policy and compliance is valuable to companies like Circle, his actions—advocating long-term confidence while personally divesting—raise questions about the alignment between his public statements and his private financial decisions, leaving investors who followed his advice to bear the market risks.

marsbit4 хв тому

Former CFTC Chairman, Circle President Tarbert: Preaching Long-Termism While Cashing Out $30 Million Himself

marsbit4 хв тому

Gate Research Institute: The 'Wall Street-ization' Wave of Crypto Financial Products – Competition or Integration?

The article titled "Gate Research Institute: Are Crypto Financial Products Sparking a 'Wall Street' Wave—Competition or Convergence?" explores the evolving relationship between the crypto ecosystem and traditional finance (TradFi). The piece begins by reflecting on Bitcoin's original 2009 vision of decentralization, disintermediation, and moving away from banks. It then contrasts this with the 2024 landscape, where key crypto assets like Bitcoin are increasingly held through Wall Street products like ETFs issued by giants like BlackRock. The article questions whether this signifies that TradFi is systematically taking over the rights to issue, price, custody, and distribute crypto financial assets. The core argument is that this is not a zero-sum takeover but rather a bidirectional convergence where each side addresses the other's weaknesses. Crypto offers 24/7 global markets, programmable settlement, and open access but lacks compliant channels, institutional-grade custody, deep fiat liquidity, and mainstream distribution. TradFi possesses these but is constrained by legacy systems, limited operating hours, and slow settlement. Two primary convergence paths are highlighted: * **Path A (CEX to TradFi):** Exemplified by Gate, which has progressed from offering tokenized stocks and CFDs to providing direct, real stock trading (US, Hong Kong, South Korea) within its platform, using USDT. * **Path B (TradFi to Crypto):** Exemplified by Robinhood, which has integrated crypto trading, acquired exchanges like Bitstamp, and is moving traditional assets like stocks onto the blockchain via tokenization and its own Layer 2. Both paths are ultimately competing to become the next-generation, unified financial account—a "super account" where users can seamlessly trade cryptocurrencies, stocks, ETFs, RWA (Real World Assets), and tokenized treasury products in one interface. The growth of RWA and tokenized treasuries (e.g., BlackRock's BUIDL) is presented as the asset-layer fusion, providing stable, yield-bearing assets on-chain and acting as a bridge between the two worlds. In conclusion, the "Wall Street-ization" of crypto is framed as a mutual transformation. Decentralized ideals persist in the protocol layer, while at the application layer, a more efficient, global, and accessible unified capital market is emerging from this convergence. The future competition lies not between crypto exchanges and stockbrokers, but between platforms vying to offer the most comprehensive asset coverage, liquidity, and user experience within a single account.

marsbit8 хв тому

Gate Research Institute: The 'Wall Street-ization' Wave of Crypto Financial Products – Competition or Integration?

marsbit8 хв тому

Claude's Major New Feature: Screen Recording + Voice, Distilling Your Skills into AI Tasks in One Click

Claude has introduced a major new feature called "Record a Skill," available for Pro, Max, and Team users. This function, found in the Claude desktop app's CoWork menu, allows users to create reusable AI skills simply by recording their screen and providing voice narration while performing a task. Claude then automatically analyzes the recording and generates a functional Skill. A hands-on test confirmed the feature works seamlessly. Users start recording via the Skills manager, perform their workflow while verbally explaining the steps and logic, and avoid including sensitive information. After recording, Claude processes the content and creates the Skill, which can be saved and later invoked with a slash command (/). This eliminates the need for manual adjustments or writing complex instruction files. The innovation goes beyond mere efficiency. Previously, creating a Skill required writing a detailed SKILL.md file in Markdown—a significant barrier for non-technical users. "Record a Skill" bypasses this by directly capturing both actions and the implicit reasoning shared in the narration. This lowers the barrier to knowledge transfer and automation, addressing a core challenge in corporate knowledge management: the difficulty of getting experts to write and maintain documentation. However, the feature also highlights a shift in the nature of work. A case study from March 2026 showed a freelancer whose five-year client relationship was effectively replaced by a hand-coded Claude Skill automating their content workflow. With the even lower barrier of screen recording, the ability to distill personal expertise into automatable skills accelerates this trend. The "moat" for work is moving from simply knowing how to do a task to mastering tasks that are difficult or impossible to automate.

marsbit13 хв тому

Claude's Major New Feature: Screen Recording + Voice, Distilling Your Skills into AI Tasks in One Click

marsbit13 хв тому

Feeding AI "Noise" Can Also Boost Scores, This Work Enables Positive Transfer with Noise

Feeding "Noise" to AI Can Improve Performance: A Method Enables Positive Transfer from Noise This work, Semi-Supervised Noise Adaptation (SSNA), introduces a Noise Adaptation Framework (NAF) that challenges traditional transfer learning. Instead of requiring a labeled source domain of real data (e.g., images, text), NAF uses randomly generated Gaussian noise as the source. For a target task with C classes, it constructs C noise clusters by sampling from Gaussian distributions. Although this synthetic noise contains no semantic meaning, NAF trains it to form a discriminative class structure in a shared representation space—clustering same-class noise and separating different classes. The key is aligning this learned structure from the noise domain to the real, sparsely labeled target domain. A small number of target labels are still essential to establish the correspondence between noise clusters and actual classes. The training objective combines: 1) supervised loss on the few labeled target samples, 2) classification loss for the noise to build its structure, and 3) a distribution alignment loss (using Negative Domain Similarity) to minimize the gap between the noise and target domains in the shared space. Experiments show significant gains in few-label settings. With just 4 labels per class, NAF with a ResNet-18 backbone improves accuracy over a standard supervised baseline (ERM) by +12.35% on CIFAR-10, +7.61% on CIFAR-100, +4.38% on DTD-47, and +2.74% on Caltech-101. It also benefits fine-grained datasets and scales to ImageNet-1K (with 100 labels/class) and text classification (AG News). NAF can be integrated into existing semi-supervised methods like FixMatch for further gains. Ablation studies confirm the transferred benefit comes from the discriminative structure of the noise, not randomness itself. Collapsing all noise into a single point causes negative transfer, while increasing separation between noise cluster centers improves performance. The amount of noise per class is less critical once a basic structure forms. In conclusion, this work demonstrates that for positive transfer, the semantic content of source data may not be necessary. What can be effectively transferred is the *organizational structure* of categories within a representation space. This offers a promising alternative for scenarios where real source data is unavailable due to privacy, copyright, or procurement constraints.

marsbit14 хв тому

Feeding AI "Noise" Can Also Boost Scores, This Work Enables Positive Transfer with Noise

marsbit14 хв тому

Торгівля

Спот
活动图片