New York Times: South Korean and Taiwanese tech companies that helped build China's hardware sector during the smartphone boom now reap the AI boom as US curbs sideline China  —  They make much of the gear that goes into giant data centers.  Demand for their products is shifting the balance of tech power.

Discussion | Link

Knicks owner James Dolan urged players to "eliminate all distractions" in a speech ahead of their playoff run. Ronald Cortes/Getty Images Before the Knicks captured the NBA title, owner James Dolan gathered the players for a rare address. In the April speech, he urged the team to "eliminate all distractions" and "sacrifice everything" in pursuit of a title. He also said that the players' families needed to understand that "they're going to have to sacrifice, too." As the New York Knicks prepared for what would become a championship run, owner James Dolan had a message for the players: Winning would require buy-in from everyone at home. The remarks came during an April 3 speech that was made public on Monday, when it was released on "The Roommates Show," a podcast hosted by Knicks players Jalen Brunson and Josh Hart. The speech marked the first time Dolan had spoken directly to the team during his tenure with the franchise. "I believe that this team can go all the way. And in my 30 years of doing this, I've never felt closer to achieving that goal than right now," Dolan said. He challenged players to fully commit to the opportunity before them, urging them to "sacrifice everything" and "eliminate all distractions" — including those at home. "I suggest that you go home and you talk to your families, you talk to your girlfriends, your wives, and you tell them about the next 10 weeks — about how you're going to be different the next 10 weeks, and how your families are going to sacrifice for the next 10 weeks," Dolan said. He then told players to make sure their families understood what would be expected of them in the weeks ahead. "Let them know what this is going to be like, what your commitment is going to be like, and how they're going to have to sacrifice, too," he said. Any concerns or problems at home can't come onto the court, he added. "For the next 10 weeks, this has just got to be all about us and about how we play." The sacrifices made by players' families would not go unnoticed, Dolan said, adding that their loved ones would also receive championship rings if the Knicks won it all. "I will buy a ring for each one of your significant others, because their contribution is going to be very important to this team winning," he said. The Knicks ended their 53-year championship drought on Saturday, when they defeated the San Antonio Spurs in Game 5 of the NBA Finals. On Thursday, the team is set to celebrate the victory with a ticker-tape parade through Lower Manhattan's famed "Canyon of Heroes." Demand for prime viewing locations has been so intense that some people have begun offering spot-holding services for a fee. The NBA championship is the latest win in a strong year for Dolan, whose net worth has grown by roughly $450 million this year to $1.9 billion, according to Bloomberg's Billionaires Index. Read the original article on Business Insider

New Fed chair Kevin Warsh is pushing to rethink how the central bank communicates with markets and the public. Tom Williams/CQ-Roll Call, Inc via Getty Images Kevin Warsh's debut as Fed chair was about more than interest rates. He unveiled plans to rethink how the central bank communicates and operates. Business leaders and analysts say the changes could reshape the markets' relationship with the Fed. The Federal Reserve held interest rates steady on Wednesday, but the bigger story was the man behind the podium. In his first meeting as Fed chair, Kevin Warsh struck a firm tone on inflation, signaling that the central bank's focus remains squarely on getting price growth back under control. Warsh also used the occasion to preview a broader shakeup at the Fed, signaling plans to rethink everything from economic forecasts and forward guidance to the institution's use of AI and internal operations. Warsh's comments suggested a more hawkish stance on inflation than many investors anticipated from Donald Trump's appointee to the Fed's top job. Here is how smart people in business are reacting to Warsh's first major policy statement. Donald Trump, US President Mandel NGAN / AFP via Getty Images President Donald Trump responded to Warsh's policies on Monday on the tarmac. "It's alright. Whatever," said Trump in response to a reporter who asked if he is aware that the Fed is holding interest rates. "It could happen," Trump added of the prospect that the Fed might even raise rates this year. "I mean, it's hard to believe. It just keeps a country down." Mohamed El-Erian, professor of practice at the Wharton School Bloomberg/Getty Images Top economist Mohamed El-Erian called Warsh "a much-needed, reform-oriented breath of fresh air" on a post on X on Monday and praised the move to create five task forces to initiate reform. "The new Federal Reserve Chair, Kevin Warsh, adopted a highly welcome change during his opening remarks at the press conference following today's policy statement," wrote El-Erian. "Underpinning his emphasis on accountability, his delivery was noticeably more open, engaging, concise, and honest," El-Erian added. Liz Thomas, chief market strategist at SoFi Warsh's first meeting as Fed chair made clear that he wants to update how the central bank operates and communicates, SoFi chief market strategist Liz Thomas wrote, pointing to his new task forces and a shorter policy statement that dropped forward guidance. "While this gives the market less to react to, perhaps it's a good thing. In my opinion, markets tend to overreact to Fed commentary and can swing widely depending on how investors interpret its messages," Thomas wrote in her weekly newsletter on LinkedIn. Even so, Thomas said the meeting was broadly hawkish. Still, she does not expect a rate hike this year, arguing that the bigger risk is that tighter policy could dampen growth. Jeffrey Gundlach, CEO of DoubleLine Capital Patrick T. Fallon / AFP via Getty Images DoubleLine Capital's Jeffrey Gundlach told CNBC on Monday that Warsh's focus on restoring price stability reduces the chances of the Fed cutting rates too aggressively and reigniting inflation. "He is absolutely telling you that he plans on delivering on price stability," said Gundlach. "We're not going to have such easy money policy as everybody thought maybe Chairman Warsh would do back in the first quarter of this year." "I think there's a greater reason to own long-term Treasuries today now that the new sheriff is in town," Gundlach said. "If you're going to get price stability, and if he doesn't deliver on something that can be characterized as price stability, he's basically announced today that he would be considered a failure." Steve Blitz, chief US economist at GlobalData.TS Lombard "Warsh gave us simple forward guidance — tighter financial conditions and less Fed in the hearts and minds of market participants," wrote Steve Blitz, the chief US economist at GlobalData.TS Lombard. Blitz said the Fed's pledge to deal with inflation was reassuring, but its willingness to let markets play a larger role in setting rates could be unsettling for traders accustomed to taking cues from policymakers. Over time, however, he argued that the approach should ultimately support productivity-led growth by leading to better capital allocation. He also suggested that Warsh's broader vision extends beyond the funds rate to shrinking the Fed's balance sheet, reducing the central bank's leverage, and encouraging the private sector to take on more. Diane Swonk, chief economist at KPMG Luke Johnson/Getty Images Diane Swonk, KPMG's chief economist, wrote on LinkedIn on Monday that Warsh has inherited the Fed's "hardest problem" — inflation. "His challenge is to restore price stability without breaking the expansion," said Swonk. "He abstained from providing a forecast for rates, but it was clear where the committee heading — nearly half in the room penciled in at least one rate hike this year and the market accurately followed suit." "We still expect two rate hikes in the back half of the year," Swonk added. Bill Adams, chief US economist at Fifth Third Commercial Bank Warsh's first meeting as Fed chair reaffirmed that the central bank remains independent of political pressure, according to Bill Adams, the chief US economist at Fifth Third Commercial Bank. "Both he and the collective weight of the rest of the FOMC used the June decision to indicate that they will hike if needed, shutting that discussion down. The Fed reaffirmed its commitment to operating independently of political pressure," Adams wrote. So while the Fed's economic outlook was largely in line with investors' expectations, Adams said markets reacted to the meeting by pricing in a greater likelihood of higher interest rates. In his view, that reaction reflected the absence of any sign that political pressure was influencing the Fed. Jeffrey Roach, chief economist at LPL Financial Warsh's meeting signaled a return to a more minimalist style of central banking communication, according to Jeffrey Roach, the chief economist at LPL Financial. "We are going back to the days of Alan Greenspan when FOMC statements were deliberately minimalist, opaque ('constructive ambiguity'), and focused on actions, not explanations," Roach wrote in a note, referring to the former Fed chair's signature cryptic communication style. Roach added that the Fed's terse Wednesday statement and closing line that "The Committee will deliver price stability" amounted to a hawkish message. Still, he argued that inflation is largely supply-driven and could ease if geopolitical tensions in the Middle East subside. Ed Yardeni, president of Yardeni Research Bloomberg/Getty Images Ed Yardeni, a longtime market vet and the president of Yardeni Research, said he was "blown away" by Warsh's press conference. "We thought he was a dove who favored lowering the federal funds rate (FFR) because he believes that AI is boosting productivity and economic growth while keeping a lid on inflation," Yardeni wrote in a note. Instead, Warsh delivered a "strict, orthodox message" on inflation and price stability. Yardeni highlighted that the Fed chair repeatedly stressed the need to return inflation to its 2% target, underscoring the central bank's focus on preventing inflation pressures from broadening. Chris Zaccarelli Chris Zaccarelli, the chief investment officer for Northlight Asset Management, said Warsh's first meeting resembled a corporate turnaround effort. "Like a skilled CEO brought in from the outside to turn a company around, the new Fed Chair reiterated his confidence in the existing leadership and practices, but organized committees and working groups to 'study' how things are currently done, setting the stage for far-reaching changes in the future," Zaccarelli wrote. The approach could lead to a less transparent Fed, changes to its inflation framework, and a smaller balance sheet, while laying the groundwork for a mix of rate cuts and balance-sheet tightening, Zaccarelli added. Read the original article on Business Insider

Federal Reserve Chair Kevin Warsh speaks to reporters during his first news conference since taking the helm at the central bank. Chip Somodevilla/Getty Images Kevin Warsh, the new Federal Reserve head, has launched task forces to assess and improve policies. Warsh's approach includes questioning assumptions and forming cross-functional review groups. His mandate included language that a CEO might use and resembled a familiar corporate strategy. Every new boss has a move. Some announce layoffs. Others reorganize the company. Kevin Warsh launched task forces. "At any institution, a change in leadership is a natural and timely opportunity to reaffirm its mission, to review current practices, and to consider whether those practices best meet our objectives," Warsh said Wednesday during a press conference. "My Fed colleagues and I will be working in close collaboration to ask what changes might improve the conduct of monetary policy." Warsh has arrived at the Fed as a new appointee of President Donald Trump, following the end of former chair Jerome Powell's tenure, with experience examining how central banks communicate. A former Federal Reserve governor, he led a 2014 review of the Bank of England's Monetary Policy Committee's transparency practices and procedures, resulting in recommendations to bolster the bank's transparency, accountability, and governance. That history makes his early focus on process and communications less surprising and suggests that his review-heavy approach may reflect a long-held interest in how central banks explain themselves, given that the Fed is likely unaccustomed to the kind of sweeping changes seen in corporate America. The mandate that Warsh laid out — to "start with first principles," "ask hard questions," and "consider alternatives" — is the kind of language a CEO might use, and resembles a familiar corporate strategy: gathering information, reassessing priorities, and building support for changes that may come later. A spokesperson for the Fed declined to comment when reached by Business Insider. How change agents operate "It's close to expected, especially when the new leader comes in with a change mandate," Mike Sacks, managing director of corporate reputation and advisory at MikeWorldWide, a strategic communications firm that advises organizations through mergers and other transitions, told Business Insider. "The first thing a new CEO usually does, especially if an outsider, is signal that things are open for debate," Sacks said. "That's partly substance, partly symbolism." To employees, a strategic review signals that leadership is reassessing whether existing practices still make sense, Sacks said. To outside stakeholders — investors in the corporate world, or financial markets in the Fed's case — it communicates that the new leader intends to put their own stamp on the institution rather than simply continue a predecessor's agenda. Warsh's creation of multiple review groups also mirrors a common change-management tactic. Leaders often establish cross-functional teams to broaden ownership of recommendations, draw on expertise from different parts of an organization, and create a structured process for evaluating potential changes, Sacks said. "You're not promising change by a certain date, you're promising a process," Sacks said. That approach leaves room for the conclusion that some areas need radical overhaul — and some may not need significant changes at all. Warsh's repeated calls to question assumptions and consider alternatives are also common in the early stages of organizational change "because it gives employees permission to surface ideas that may not have moved under the prior leadership structure," Sacks said. Still, he cautioned that there are limits to that message. If leaders focus too heavily on change, they risk undermining institutional memory and expertise. "A good version of this language is not 'everything is up for grabs,'" Sacks said. "It is: 'We respect the institution's expertise, but no assumption is beyond review.'" A signal of changes ahead Jo-Ellen Pozner, an associate professor of management and entrepreneurship at Santa Clara University's Leavey School of Business, said Warsh's approach "does feel corporate." Leaders, including CEOs, are wise to take the time to understand the ins and outs of an organization, including its advantages and dysfunctions, before announcing sweeping changes, she said. A task force could be a way to bring about big changes, she said. "It does seem like it's intentionally meant to shake things up and to signal change and the desire for change pretty clearly," Pozner said. Pozner said that at the Fed, staffers tend to be professionals "with deep training and experience." For highly experienced staff like that, she said, big changes announced by leadership can rankle those who have been trained over many years to think and act a certain way. "The danger is that you lose the people who are most committed to the initial vision, and the Fed is meant to be an apolitical, stable institution," she said. The Fed chair's job, like any leader, is to corral decision-making and ensure that everybody is aligned around the right set of priorities. For the Fed, that means focusing on its dual mandate of balancing maximum employment and inflation — but that doesn't mean the chair shouldn't also be concerned with other efforts, such as reducing waste and improving the organization's culture, Pozner said. Jerome Powell, former chairman of the US Federal Reserve, departed the role in May. Bloomberg/Getty Images The classic corporate playbook Sacks said many of the management techniques Warsh is employing — strategic reviews, internal workstreams, structured decision-making, and testing assumptions — are commonly used in both public and private institutions. However, there are important differences. While CEOs can justify changes by pointing to shareholder value or competitive performance, Sacks said, a Fed chair operates under a mandate of institutional independence and faces far greater sensitivity to political interpretation. "The same management and communications technique that looks decisive in a company can look more complicated in this context," Sacks said. Taken together, the moves resemble a familiar phase in the life cycle of a large organization, Sacks said. "This sounds like the early stage of an organizational reset, often what you see in a new leader's first 100 days," he said. "The leader has not yet announced the full change agenda, but they are creating the conditions for one: reviewing strategy, testing assumptions, forming working groups, and building internal legitimacy." That doesn't necessarily mean sweeping changes are on the way, Sacks added. Sometimes, the review process serves a different purpose: helping a new leader determine what should stay exactly as it is. Read the original article on Business Insider

The egg sandwich showdown: Lawson, 7-Eleven, or Family Mart? Cheryl Teh Egg sandwiches are a popular comfort food in Japanese convenience stores, or konbinis. At Family Mart, Lawson, and 7-Eleven, you can get one for less than $2. I, a lover of egg salad sandwiches, set out to rank them on my trip to Tokyo this summer. I've been to Japan over a dozen times and eaten enough convenience store egg salad sandwiches to call myself an authority. I've had them for breakfast, at the standing tables at a 7-Eleven in Osaka. I've had them for lunch, with a slice of Family Mart chicken stuffed in the middle, a budget traveler's prime source of sustenance. And I've eaten them at 2 a.m., washed down with ice-cold Ito-En green tea straight out of a hotel room minibar. I am not alone in my love for egg sandos. On my latest trip, foreigners like myself were cleaning out the shelves at Tokyo konbinis at lunch time. Even the late food great Anthony Bourdain lauded the humble Lawson egg sando for its "unnatural, inexplicable deliciousness." If you're wondering why I — and legions of foreigners — were rooting around convenience stores, these shops aren't anything like their US counterparts. Japanese convenience stores sell everything from full, reheatable meals to exclusive merchandise, like Muji socks and special konbini-only makeup lines. The stores went viral during the 2021 Olympic Games, when athletes and journalists happened upon 24-hour outlets that stocked, seemingly, everything they needed. It's big business, too. According to data released by the Japan Franchise Association, sales at Japanese convenience stores hit a high in 2025, breaking records for the fourth straight year. All-store sales went up 2.2% from the year before, hitting 12.05 trillion Japanese yen, or $75.8 billion. This summer, I went back to Tokyo and embarked on a mission: to give you the down low on one of Japan's cheapest but most comforting (in my humble opinion) convenience store offerings. I went to Family Mart, Lawson, and 7-Eleven, and chowed down. Lawson Anthony Bourdain had nothing but praise for the humble Lawson egg sando. Cheryl Teh The egg sando, or tamago sando, isn't just your plain old sandwich. The Japanese milk bread hits different, the mayonnaise is light, and the savory texture of the egg salad makes for a great, creamy filling. For a quick bite on the go, you'd be hard pressed to find a better option. In terms of price, all three chains sell their sandwiches at just under 300 Japanese yen, or around $1.85. Maybe I was hungrier than usual — I'd grabbed the Lawson sandwich as a to-go meal to munch on outside the Tokyo Dome, while waiting for the Seventeen concert to start. Or maybe the sandwich was just that good. "Pillows of love," Bourdain called these sandwiches. And he was right — the Lawson egg sandwich was perfection in plastic wrap. The shokupan was soft, milky, and just sweet enough without being cloying. And the egg, even an hour out of the cooler, was absurdly tasty. Family Mart I believe Family Mart egg sandos are best consumed with a slice of hot Famichiki — the chain's store-brand chicken cutlet. Cheryl Teh Family Mart, in my opinion, has the best chicken cutlets among the Japanese konbini stores — another debate, for another time. This time, I had to evaluate the merits of the egg sando as a standalone. I found Family Mart's bread a little thinner and dryer than the 7-Eleven and Lawson options. The egg mayonnaise, however, was savory and delectable. 7-Eleven The 7-11 sandwich was a great midnight snack. Cheryl Teh Where the Family Mart sandwich lost out in the bread wars, 7-Eleven's lost in the egg fight. I had not one, but three 7-Eleven sandwiches during my weeklong trip in Tokyo, mostly because my hotel was next to an outlet. The bread was great: spongy, perfectly cut, and still at a perfect consistency even when consumed as a late-night snack — but the eggs were not as hearty as its competitors' offerings. My friend disagreed: 7-Eleven is her favorite, and its lighter, less-salty eggs were far more appealing to her. The verdict It was a hard fight — and I do like all three egg sandos, and would gladly have one any day. Cheryl Teh Lawson took top billing, with Family Mart and 7-Eleven tied for second. I don't recommend that you eat egg sandos for every meal in Japan — because, come on, you're in Japan. I've had some great lunches and dinners in the country: a kaisendon on Enoshima Island with a view of Mt. Fuji; wagyu chuck flap tail on the grill; omakase and shabu-shabu at TikTok-viral locales that we had to book months in advance. But some of my best memories of traveling in Japan have involved me, my friends, and our egg sandos — eaten, with gusto, during 6 a.m. Shinkansen rides, or at post-karaoke supper. Read the original article on Business Insider

arXiv:2606.18406v1 Announce Type: new Abstract: Personalized dialogue agents require continuous long-term memory to maintain coherent interactions across multiple sessions. However, deploying these capabilities on consumer-grade hardware (e.g., 8 GB VRAM edge devices) introduces severe memory and compute bottlenecks. Existing systems typically rely on isotropic cosine similarity for retrieval and heuristic rules for context compression. These approaches lack a unified theoretical foundation, frequently suffering from the hubness problem in high-dimensional retrieval and syntactic fragmentation during compression. To overcome these limitations, we propose CoreMem, a resource-efficient edge-cloud memory architecture fundamentally unified by information geometry. First, Riemannian retrieval replaces cosine matching with a locally adaptive Fisher-Rao metric, effectively penalizing hub memories via Mahalanobis distance with O(Ndr) Woodbury acceleration for real-time search. Second, Fisher-guided discrete token distillation (FDTD) introduces a hierarchical sentence-to-token compression mechanism. It derives sensitivity scores from Fisher information traces, providing a principled compression-KL tradeoff augmented with explicit structural syntax protection. Evaluated on the LOCOMO and LongMemEval-S benchmarks, CoreMem achieves strong accuracy improvements, yielding substantial gains in Open-domain (+4.51 pp) and Temporal (+4.17 pp) reasoning. Extensive profiling confirms that CoreMem operates seamlessly within a strict 8 GB VRAM budget, successfully bridging the gap between resource-constrained edge devices and the demand for theoretically grounded, lifelong memory agents.

arXiv:2606.18372v1 Announce Type: new Abstract: Educational dialogue is a valuable but sensitive resource for research: the same transcripts that capture authentic learning often capture personally identifiable information (PII) entangled with curricular content, where "Riemann" may refer to a real student or to a mathematical concept. Existing approaches force a tradeoff between governance and accuracy. Commercial Large Language Models (LLMs) can handle this ambiguity but require sending student data to third parties, while local named entity recognition (NER) systems preserve governance but over-redact curricular terms. We propose a fully local cascade framework that reframes de-identification from open-ended entity recognition to constrained privacy triage. A recall-first union proposer combines two lightweight encoders with deterministic rules to over-generate candidate spans; a context-aware reviewer then makes a binary Redact/Keep decision for each candidate using surrounding dialogue and speaker role. We evaluate three reviewer configurations against same-family LLM-only baselines and a commercial API on math tutoring transcripts from two large platforms. The strongest local configuration reaches 0.958 macro F1, compared with 0.767 for a same-family LLM-only baseline and 0.706 for the commercial API, while running entirely on a single laptop. On a targeted challenge set of curricular-personal name ambiguity, the same configuration degrades by only 0.03 F1 versus 0.19 to 0.25 for smaller reviewers. These results suggest that for educational de-identification, problem formulation matters more than model scale.

arXiv:2606.17188v2 Announce Type: new Abstract: Current multilingual evaluations for Vision-Language Models (VLMs) assume a one-to-one mapping between language and orthography, overlooking billions of users of multi-script languages. We introduce PuMVR (Punjabi Multimodal Visual Reasoning), a benchmark of 1,000 strictly parallel image-text instances across Punjabi's three active scripts: Gurmukhi, Shahmukhi, and Roman. Evaluating 10 state-of-the-art VLMs, we expose a substantial and systematic Script Gap. Models frequently solve visual tasks in one script while failing identical tasks in another, with accuracy deltas reaching 16%. Crucially, visual input boosts absolute performance uniformly yet does not close the orthographic gap. Furthermore, cross-script in-context transfer is highly brittle, exposing script-locked knowledge representation. Supported by McNemar tests across all script pairs, our findings demonstrate that current "multilingual" VLMs are not truly multi-script. We propose the Script Consistency Rate (SCR), which falls as low as 24.8% on our benchmark, as a mandatory metric for script-agnostic evaluation to ensure equitable AI access. Data and code are available at: https://github.com/prabhjotschugh/Not-Truly-Multilingual-PuMVR.

arXiv:2606.18273v1 Announce Type: new Abstract: Large audio language models (LALMs) have shown impressive capabilities on diverse audio understanding tasks, ranging from speech transcription to music analysis. However, because LALMs are typically trained to produce text-aligned responses, their hidden states are progressively shaped for text generation rather than for preserving acoustic information. As a result, the diverse acoustic content that audio carries, such as phonetic detail, prosody, sound events, affect, and pitch, is lost along the way and difficult to leverage in the response. We introduce Continuous Audio Thinking (CoAT), a framework that equips audio language models with a continuous latent workspace for organizing acoustic information prior to response generation, grounded by distillation from audio experts. Within the thinking space, the model can utilize the rich acoustic information provided by expert distillation when generating its response. Furthermore, the proposed continuous thinking block can be processed in a single prefill, so CoAT does not require additional autoregressive decoding cost over the baseline. Across three LALMs, Qwen2-Audio, Qwen2.5-Omni-7B, and Audio Flamingo~3, performance gains on a broad benchmark suite spanning audio reasoning, audio understanding, music classification, speech emotion, and speech transcription demonstrate the effectiveness of CoAT. Further analysis confirms that the auxiliary supervision propagates from the thinking positions to the model's textual responses.

arXiv:2606.18287v1 Announce Type: new Abstract: Multimodal neuroimaging, integrating functional connectivity from fMRI and structural connectivity from DTI, enables non-invasive analysis of brain networks using graph neural networks. However, demographic factors such as age and sex systematically confound the relationship between brain connectivity and clinical outcomes, causing GNNs to exploit spurious shortcuts rather than learning causally invariant representations. While recent causal GNN methods introduce causality at the graph-modeling level, their causal mechanisms remain domain-agnostic without accounting for the real-world confounders inherent in clinical neuroimaging data. Moreover, brain networks are constructed from atlas-based parcellations where each region exhibits distinct sensitivity to demographic factors, necessitating region-aware adjustment. We propose Artemis, a region-level causal framework that bridges this gap with causal intervention at each brain region independently by learning region-specific confounder representations with lightweight parameters. Our adjustment comprehensively utilized the multimodal functional and structural features for graph reasoning as a plug-in module compatible with arbitrary GNN backbones. Experiments on three benchmarks, ADNI for disease diagnosis, OASIS for dementia staging, and HCP for sex classification, demonstrate consistent improvements over representative GNN-based baselines. Multiple supporting experiments further demonstrate statistical significance and neuroscientific interpretability.

arXiv:2606.17246v1 Announce Type: new Abstract: Remote-sensing vision-language models (RS-VLMs) have advanced Earth-observation analysis toward visual interpretation and instruction-following, yet fall short of operational geo-intelligence, which demands tool-grounded spatial reasoning and structured, evidence-backed decisions. We introduce GeoDisaster, an operational geospatial disaster reasoning benchmark with 2,921 verified instances across 43 question types and five task families: deforestation monitoring, multi-hazard analysis, building-damage assessment, flood-safe routing, and Sentinel-1 SAR flood monitoring. Instances integrate heterogeneous EO/GIS evidence-optical and SAR imagery, raster masks, vector geometries, road networks, and exposure layers-spanning hazard detection, damage assessment, exposure estimation, and diagnostic report generation. Ground-truth answers are grounded in executable geospatial workflows and deterministic consistency checks, removing the need for language-model annotation. We further propose an orchestrated multi-agent framework with 18 disaster-oriented tools, where role-specialized agents coordinate through explicit execution contracts, aligned via Role-Contract Expectation Alignment (RCEA): failure-aware supervised fine-tuning combined with contract-grounded reinforcement learning over dense step-level signals. Experiments show that GeoDisaster challenges existing RS-VLMs and agentic systems, while RCEA improves tool use, evidence grounding, state consistency, and decision generation.

arXiv:2606.17242v1 Announce Type: new Abstract: Coastal algal bloom monitoring requires frequent, spatially detailed, and globally consistent observations, provided by Landsat-8/9 and Sentinel-2 A/B/C. Together, these missions offer over a decade of medium-resolution multispectral imagery with near-global coverage every 2-3 days, enabling the detection of fragmented bloom structures not resolvable by coarse ocean-color sensors. However, their use in aquatic environments remains challenging due to limited spectral coverage and a lack of harmonized reflectance products. As an alternative to traditional bio-optical methods, deep learning-based image classification offers a data-driven approach that can overcome many of these limitations. This study presents the first successful implementation of vision transformer-based coastal algal bloom mapping using 30-m Landsat-Sentinel-2 images. A globally distributed bloom patch dataset was generated across bloom-prone coastal hotspots worldwide. Four transformer-based architectures were compared against a standard convolutional baseline for fine-scale bloom detection, and assessed under different optical water types and atmospheric and surface conditions. All deep learning models showed strong capabilities in detecting floating bloom areas, with omission and commission errors of 8-65%. Under cloud and glint stress in a time series, the Swin Transformer outperformed traditional spectral-index approaches, which produced widespread false positives, effectively avoiding cloud- and glint-affected pixels. Comparisons with MODIS-derived products further highlighted the benefits of higher spatial resolution in detecting fragmented and irregularly affected blooms. Our findings support deep learning as a reliable tool for medium-resolution, consistent monitoring of floating algal blooms in dynamic coastal environments.

arXiv:2606.18448v1 Announce Type: new Abstract: Computer-use agents (CUAs) approach human-level performance on standardised benchmarks but still struggle on long-horizon tasks and unseen software. Existing skill libraries address this with reusable skills, but represent the skill artifact as text only, despite the visual nature of GUI interaction. We propose VISUALSKILL: a hierarchical multimodal skill, tailored to each target application and organised as a central index over per-topic files, which the agent consumes through a load_topic MCP tool that fetches the relevant topic's text and figures on demand. We construct each skill with a two-stage pipeline that combines authored documentation with live-application UI exploration. On two CUA benchmarks, CUA-World and OSExpert-Eval, a Claude Code CLI agent backed by Claude Opus 4.6 reaches an average score of 0.456 with VISUALSKILL, a +15.3 point absolute lift over the no-skill baseline (0.303). Against a matched text-only skill that is generated from the same source content and differs from VISUALSKILL only in modality, VISUALSKILL yields a further +8.3 point absolute gain over the matched text-only skill (0.373 vs. 0.456), providing direct evidence that retaining visual figures in the skill artifact, rather than verbalizing them away, helps the agent both identify UI elements and verify workflow state after each action. Our code is available at https://github.com/XMHZZ2018/VisualSkills.

arXiv:2606.17296v1 Announce Type: new Abstract: Unified multimodal models (UMMs) have recently emerged as a promising paradigm for integrating multimodal understanding and generation within a single autoregressive transformer. However, during multimodal instruction tuning, these models often exhibit pronounced modality imbalance: language gradients dominate optimization, thus leading to lower image generation quality, especially under parameter-efficient fine-tuning such as LoRA. In this work, we systematically analyze modality imbalance in LoRA-based fine-tuning of UMMs for interleaved text-image generation. We show that vision modality performance degrades substantially more than text modality performance when compared to unimodal counterparts, and that modality-specific gradients can differ by orders of magnitude across various tasks and layers. Motivated by this observation, we reformulate the multimodal instruction tuning as a bi-objective optimization problem and propose Pareto LoRA, a Pareto-optimal gradient integration strategy that balances the text and image objectives by modulating the gradient direction and strength. Experiments on the CoMM benchmark with Emu2 demonstrate that Pareto LoRA consistently improves multimodal generation balance, achieving up to 44.9% gains in perceptual image quality over vanilla LoRA while maintaining comparable text performance.

arXiv:2606.17279v1 Announce Type: new Abstract: Surgical video question answering requires multi-step reasoning across semantic, spatial, and temporal dimensions. Existing methods architecturally compress videos into discrete token representations and couple visual perception with reasoning. This approach fragments continuous spatial-temporal relationships and has been shown to restrict multi-step reasoning capabilities. We introduce a reinforcement learning (RL) framework that trains large language models (LLMs) to decouple perception from reasoning by operating over digital twin representations constructed from surgical foundation models. Additionally, we introduce hierarchical representations across frame, temporal window, and procedure levels with probabilistic uncertainty estimates. Finally, we propose a novel reward that combines format validation with accuracy assessment through clinical plausibility evaluation and uncertainty-aware calibration for training. To demonstrate the capabilities of this approach, we introduce REAL-Colon-Reason, a colonoscopic benchmark with 2000 question-answer pairs across three complexity levels. We achieve state-of-the-art performance on REAL-Colon-Reason and two existing surgical VideoQA benchmarks REAL-Colon-VQA and EndoVis18-VQA.

arXiv:2606.17241v1 Announce Type: new Abstract: Continuous AI inference on resource-constrained edge hardware introduces deployment effects that are largely invisible to conventional benchmark evaluation, including temporal instability in streaming video, thermal throttling under sustained load, and workload-dependent performance variability. We present Edge-TSR, a deployment-oriented continuous edge inference system for sustained roadside perception on the NVIDIA Jetson Orin Nano. Edge-TSR integrates detection, tracking, fine-grained classification, and a lightweight track-aware temporal stabilization mechanism that improves streaming inference consistency with negligible computational overhead. Our central finding is that benchmark-centric evaluation systematically overstates deployed edge inference performance. Across three state-of-the-art baselines, we observe consistent 20-30% relative degradation when transitioning from static-image evaluation to real-world streaming deployment. Edge-TSR addresses this gap through temporal inference stabilization, recovering up to 10.16% classification accuracy over per-frame inference baselines while maintaining sustained real-time performance under continuous operation. We evaluate the complete system under diverse real-world deployment conditions, jointly characterizing inference quality, latency, throughput, and thermal behavior during long-duration operation. A 55-minute vehicular deployment over a 26 km route demonstrates sustained operation at 16.18 FPS within safe thermal limits on a single embedded device without cloud offload. Our findings show that deployment-aware evaluation and temporal inference stabilization are necessary components of continuously operating edge AI systems intended for real-world sensing deployments. We release a sample annotated streaming video evaluation dataset and full system implementation to support reproducible deployment-centric evaluation.

arXiv:2606.17222v1 Announce Type: new Abstract: Hyperspectral image (HSI) crop analysis is essential for precision agriculture because it captures rich spectral and spatial information for accurate crop monitoring and assessment. However, HSI classification remains challenging due to high spectral dimensionality, spatial complexity, class imbalance, and limited labeled samples. To address these challenges, this paper proposes a BiSpectral Mamba-based framework that combines multi-scale convolutional feature extraction, spectral attention, bidirectional state-space modeling, and quantum-inspired learning. A multi-scale CNN backbone first extracts hierarchical spatial-spectral representations through feature fusion across multiple resolutions. A spectral attention mechanism then emphasizes informative bands while suppressing redundant and noisy channels. The refined features are processed by a BiSpectral Mamba module that captures long-range dependencies in both forward and backward directions by modeling hyperspectral feature maps as sequential tokens. In addition, class-weighted optimization and feature fusion strategies are incorporated to improve training stability and mitigate class imbalance. Experimental evaluation on the UAVHSI-Crop dataset demonstrates the effectiveness of the proposed framework, achieving an overall accuracy of 84.83%. The results show that integrating convolutional, attention-based, and state-space modeling components enables robust spatial-spectral feature learning for crop classification. The proposed framework also shows potential for broader agricultural and remote sensing applications, including crop disease detection, yield prediction, and soil moisture estimation, while highlighting the effectiveness of structured state-space and quantum-inspired architectures for hyperspectral image analysis.

arXiv:2606.18453v1 Announce Type: new Abstract: Large language models (LLMs) exhibit substantial cross-lingual variation in mathematical reasoning performance, but it remains unclear whether these differences reflect language-specific parameters or a shared mechanism that manifests differently by language. We present a cross-lingual mechanistic analysis of mathematical reasoning in LLMs, enabling us to localize and compare model parameters that support mathematical reasoning across languages. We find that the extracted math-associated parameters exhibit partial cross-lingual overlap, with the strongest overlap concentrated in intermediate model layers. We further observe that English consistently produces the largest set of math-relevant parameters, whereas lower-resource languages reveal smaller sets of relevant parameters. These results suggest that math-related behavior in multilingual LLMs is neither fully language-invariant nor fully language-specific, but instead exhibits partial cross-lingual parameter overlap with systematic language-dependent differences.

arXiv:2606.18394v1 Announce Type: new Abstract: Speculative decoding (SD) accelerates autoregressive Large Language Models (LLMs) by drafting multiple tokens and verifying them in parallel, but it faces a scaling limitation: increasing the draft budget improves speed only when acceptance remains high and drafting overhead stays low. This ceiling has been difficult to break because prior head-based SD methods face a causality-efficiency dilemma. Autoregressive drafters produce path-conditioned candidates that are effective for tree speculative decoding with higher acceptance length, but their drafting cost grows with tree depth. Bidirectional block-diffusion drafters generate all positions in one pass, but their branch-agnostic marginals can form individually plausible yet mutually inconsistent trees, wasting budget and reducing acceptance. We propose JetFlow, a head-based SD framework that combines one-forward drafting efficiency with branch-wise causal conditioning. JetFlow trains a causal parallel draft head over fused hidden states from the frozen target model, producing candidate trees whose scores align with the target model's autoregressive factorization. This enables JetFlow to convert larger draft budgets into longer accepted prefixes and higher end-to-end speedup. Across math, coding, and chat benchmarks on dense and MoE Qwen3 models, JetFlow consistently outperforms bidirectional-head and tree-based SD baselines. On H100 GPUs, JetFlow achieves up to 9.64x speedup on MATH-500 and 4.58x on open-ended conversational workloads, with further latency gains demonstrated through vLLM integration under realistic serving loads. Our code and models are available at https://github.com/hao-ai-lab/JetFlow.